Open any introductory psychology textbook published in the past forty years and turn to the chapter on emotion. Somewhere in the first ten pages you will find the story.

Stanley Schachter and Jerome Singer, the textbook will say, settled a long-running debate about the nature of emotion. They injected subjects with epinephrine — adrenaline — and told them it was a vitamin compound called “Suproxin.” Then they put each subject in a waiting room with a confederate, an actor pretending to be another subject. Sometimes the confederate acted euphoric, throwing paper airplanes, dancing, hula-hooping. Sometimes the confederate acted angry, ripping up a questionnaire and storming out. The result, the textbook will say, was clean and elegant. Subjects injected with adrenaline and exposed to the happy confederate felt euphoric. Subjects injected with adrenaline and exposed to the angry confederate felt angry. Same physiological arousal. Different cognitive context. Different emotion.

The lesson the textbook draws is that emotion is not raw physiology and not pure cognition. Emotion is a label we attach to undifferentiated arousal, and the label comes from whatever is going on around us. This is the “two-factor theory of emotion,” and it has been the canonical account of how emotions work — at least the version taught to undergraduates — since the paper was published in Psychological Review in 1962.

It also, in its strict experimental form, has never cleanly replicated.

When Gary Marshall and Philip Zimbardo set out in 1979 to reproduce the euphoria condition, the effect did not appear. When Christina Maslach set out the same year to reproduce the anger condition using a cleaner design, unexplained arousal consistently produced negative emotions regardless of what the confederate was doing — the opposite of what the theory predicts. When Rainer Reisenzein conducted a comprehensive review of twenty years of follow-up work in 1983, he concluded that two of the three central predictions of the theory were “not adequately supported by the data,” and that the version of the theory that survived was a substantially attenuated one.

The textbook treatment, in other words, is about four decades out of date. The point of this article is not to discredit Schachter and Singer — their broader intuition that cognition shapes emotional experience has held up in modern theories — but to be precise about what the empirical record actually supports and what it does not, because the difference matters for anyone trying to use this work as a foundation for “emotional intelligence” training, sales psychology, marketing arousal claims, or leadership development. The lab evidence for the strict mechanism is weaker than the popular framing implies. That is a story worth knowing.

What Schachter & Singer 1962 Actually Tested

The experiment is famous, often retold, and usually retold incorrectly. Here is what actually happened.

Schachter and Singer recruited 185 male students from the introductory psychology course at the University of Minnesota. Each subject came to the lab one at a time and was told the study was about the effects of a vitamin compound called “Suproxin” on vision. After signing a consent form for the injection, each subject was given one of two injections: epinephrine (adrenaline), which produces physiological arousal — increased heart rate, flushing, tremor, faster breathing — within roughly three to five minutes; or a saline placebo, which produces no physiological change.

Subjects in the epinephrine condition were further split into three sub-conditions based on what they were told about the side effects:

  • Epinephrine Informed: told to expect the actual symptoms (palpitations, hand tremor, warm face).
  • Epinephrine Ignorant: told nothing about the symptoms.
  • Epinephrine Misinformed: told to expect different symptoms (itching, headache, numbness).

The theoretical logic was that the Informed group had a ready explanation for their arousal (“the shot is doing it”) and so would not need to search the environment for an explanation. The Ignorant and Misinformed groups would have no explanation for their bodily state, would search the environment for an interpretation, and would therefore be susceptible to whatever emotional cues were present.

After the injection, each subject was led to a waiting room and seated with another “subject” — actually a confederate — to wait for the vision tests to begin. The confederate enacted one of two scripts.

In the euphoria condition, the confederate doodled on paper, then crumpled the paper into a basketball and tried to throw it into the wastebasket, then built paper airplanes and threw them, then constructed a tower of folders, then twirled a hula hoop. The script was designed to escalate into goofy, expressive happiness, and the confederate was instructed to try to draw the subject into participating.

In the anger condition, the subject and the confederate were both given a long, increasingly insulting questionnaire to fill out — asking, for example, how many men in the subject’s family weighed over 200 pounds and what the subject’s mother’s extramarital affairs had been. The confederate complained about the questions in escalating language, eventually tore up the questionnaire, threw it on the floor, and stormed out of the room.

Two kinds of dependent measures were collected. Observational data: experimenters watched through a one-way mirror and coded whether the subject joined in the euphoric play (threw the basketball, flew an airplane, used the hula hoop) or matched the angry behavior (made angry comments, agreed with the confederate’s complaints). Self-report data: after the waiting room interaction, subjects filled out a questionnaire asking how angry, irritated, happy, or good they felt.

The reported pattern of results, in the most-cited summary version of the paper, was that Epinephrine Ignorant and Epinephrine Misinformed subjects showed more emotional behavior (matching the confederate) than Epinephrine Informed subjects. Subjects in the placebo condition fell somewhere in between. (Schachter & Singer, 1962, DOI 10.1037/h0046234)

This is the result that made it into the textbooks.

The Methodological Issues That Were Already Apparent In 1962

Even by the publication standards of 1962 — which were far looser than today’s — the original paper had problems that careful readers noticed at the time.

Post-hoc subject exclusions. A nontrivial number of subjects were excluded from the analysis. In the euphoria condition, five subjects were excluded for being suspicious of the confederate. In the anger condition, eleven subjects were excluded because they considered the insulting questionnaire either fair or amusing rather than infuriating. The final cell sizes after exclusions were small: roughly 20 to 30 subjects per condition in the original between-subjects design. When researchers later asked whether the pattern of results would survive if the excluded subjects were included, the effect attenuated substantially.

Weak and inconsistent statistical patterns. Even on the reported data, several of the predicted contrasts were not statistically significant. On the self-report measures of euphoria, the differences between groups were small and in some cases failed to reach conventional significance. The behavioral measures showed a clearer pattern, but the behavioral measures were also more subjective — observers coding behavior through a one-way mirror cannot be fully blind to condition, and the coding required interpretation. A 1962 follow-up by Schachter and Wheeler in fact reported that observer ratings of amusement, while in the predicted direction, were “not significant,” with film ratings only at borderline significance — a result the authors themselves acknowledged.

Ambiguous emotion measurement. The dependent measures conflated two different things: the subjective experience of an emotion (“I felt angry”) and the behavioral expression of an emotion (matching what the confederate was doing). A subject who joins a confederate in throwing paper airplanes may be doing so out of euphoria, out of social conformity, out of relief at being asked to do something other than wait, or simply because the confederate is harder to ignore than the wall. The design did not distinguish these possibilities.

Confounded “epinephrine” manipulation. The epinephrine dose was a fixed 0.5cc of 1:1000 epinephrine solution, regardless of body weight, age, or baseline arousal. Some subjects experienced strong physiological effects, others felt almost nothing. The “arousal” condition was not really uniform across subjects, and the analysis did not control for actual physiological response.

These issues did not invalidate the work — early-career social psychology in the 1960s was full of similar designs, and the paper’s contribution was conceptual as much as empirical — but they meant that the empirical foundation under the textbook story was thinner than the textbook story suggested. The replication failures that followed seventeen years later were not a surprise to anyone who had read the original paper carefully.

The 1979 Replication Crash

In 1979, two independent research groups published replication attempts in the same issue of the Journal of Personality and Social Psychology. Neither reproduced the original finding.

Marshall and Zimbardo: the euphoria condition. Gary Marshall and Philip Zimbardo (yes, the same Philip Zimbardo of the Stanford Prison Experiment) ran a careful, well-powered replication of the euphoria condition. They used a larger sample, a refined version of the confederate script, and improved physiological monitoring. The result: epinephrine-injected subjects exposed to the euphoric confederate did not feel more euphoric than placebo subjects. The confederate’s behavior produced essentially no effect on the subjects’ emotional state — happy, sad, or otherwise. To the extent epinephrine-injected subjects reported any consistent emotional change relative to placebo, it was a slight tendency toward negative affect, not euphoria.

Marshall and Zimbardo concluded that “epinephrine-injected subjects were not particularly susceptible to the euphoric confederate’s emotional contagion” and that the original Schachter and Singer pattern, for the euphoria condition, did not appear to be reliable. (Marshall & Zimbardo, 1979, DOI 10.1037/0022-3514.37.6.970)

Maslach: the anger condition. Christina Maslach (later famous for her work on burnout) took a different approach to the same question. Rather than use epinephrine, which has variable effects across subjects, she induced arousal through hypnotic suggestion — a method that produces more uniform physiological responses. She crossed this with euphoric and angry confederate conditions modeled on the original Schachter and Singer design.

The result was even more damaging to the original theory than Marshall and Zimbardo’s. When subjects experienced unexplained arousal, they consistently reported negative emotional states — anger, irritation, anxiety, discomfort — regardless of whether the confederate was acting euphoric or angry. Unexplained arousal did not get neutrally labeled by environmental context. It got labeled negatively, as something unpleasant happening to the body, and the confederate’s behavior did not redirect it toward positive emotions.

Maslach concluded that “undifferentiated arousal, in the absence of a clear explanation, tends to be interpreted as inherently negative” — a result that directly contradicts the two-factor theory’s central claim that the arousal can be relabeled as anything depending on cognitive context. (Maslach, 1979, DOI 10.1037/0022-3514.37.6.953)

Two careful 1979 replications, two failures to reproduce the original finding, both published in the field’s top social psychology journal in the same issue. This was not a quiet disconfirmation. It was a coordinated, public, methodologically careful refutation of the strict version of the theory. And yet the textbooks did not change.

Reisenzein’s 1983 Comprehensive Review

Four years after the 1979 replications, Rainer Reisenzein published in Psychological Bulletin what remains the most thorough single review of the two decades of follow-up work to Schachter and Singer 1962. Reisenzein decomposed the theory into three central testable predictions and evaluated the evidence for each:

  1. Arousal is necessary for emotion. The claim that without physiological arousal, an emotional state cannot occur.
  2. Unexplained arousal gets labeled by environmental cues. The core mechanism — same arousal becomes different emotions depending on context.
  3. Misattributed arousal from an extraneous source intensifies emotional reactions. A weaker corollary — arousal from one source, attributed to another, can amplify the second emotion.

Reisenzein’s verdict, after reviewing roughly fifty published studies spanning 1962 to 1983:

  • The first prediction — arousal as necessary for emotion — had no convincing empirical support. Emotional states can occur without measurable physiological arousal, and many cases of strong physiological arousal produce no specific emotional state at all.
  • The second prediction — the core labeling mechanism — was also not supported. The bulk of well-designed studies failed to find the predicted effect. Unexplained arousal does not appear to be a blank slate that environmental cues can write any emotion onto.
  • The third prediction — misattribution intensifies — was the only one with reasonable empirical support. Arousal from one source, attributed to a different emotional cause, can amplify the second emotion. But this is a much weaker claim than the original theory, and it specifies a narrower mechanism than the textbook version.

Reisenzein’s overall summary: “The role of arousal in emotion has been overstated, and the available data support at best a rather attenuated version of Schachter’s theory — that arousal feedback can have an intensifying effect on emotional states — and that this arousal-emotion relationship is mediated, in part, by causal attributions regarding the source of arousal.” (Reisenzein, 1983, DOI 10.1037/0033-2909.94.2.239)

This is a careful, scholarly verdict. It does not say the work was wrong or fraudulent. It says the strong form of the theory — the form that became canonical in textbooks — is not what the evidence supports. The attenuated form that survives is narrower, more conditional, and explains far less than the textbook version claims.

A separate earlier review by John Cotton in 1981 in the European Journal of Social Psychology reached compatible conclusions, also finding the support mixed at best. (Cotton, 1981, DOI 10.1002/ejsp.2420110402)

What Survives In Modern Emotion Research

The honest summary of where things stand now, in 2026, is that the two-factor theory of emotion — in its strict 1962 form — is not what most active emotion researchers believe. But Schachter and Singer’s broader intuition, that cognition and context shape emotional experience, has been absorbed into the modern theories that replaced strict two-factor theory.

Russell’s circumplex model of affect (1980 and after) treats emotion as varying along two continuous dimensions — valence (pleasant–unpleasant) and arousal (activated–deactivated) — with discrete emotions emerging as conceptual categories applied to specific regions of that space. This preserves the spirit of two-factor theory without committing to the specific lab paradigm that failed to replicate.

Carver and Harmon-Jones (2009), in a major Psychological Bulletin reassessment of anger as an emotion, explicitly note that the original Schachter-Singer story does not fit the modern evidence. Anger, for example, is an “approach-related” emotion that activates left-frontal cortical regions associated with goal-pursuit, not a generalized arousal pattern that could equally be labeled euphoria. The neurology of distinct emotions turns out to involve distinct circuits, not a single undifferentiated arousal substrate. (Carver & Harmon-Jones, 2009, DOI 10.1037/a0013965)

Lisa Feldman Barrett’s theory of constructed emotion (2017 book, plus a 2025 multi-author update in Perspectives on Psychological Science) goes substantially further. In Barrett’s view, emotions are not labels applied to arousal — they are concepts constructed in the brain from interoceptive predictions, sensory input, cultural categories, and learned associations. The brain is not interpreting raw arousal; it is generating an emotional experience predictively, based on what it expects given the context. This is a much more sophisticated theory than two-factor, draws on a much larger evidence base in cognitive neuroscience, and treats the cultural and linguistic shaping of emotion as central rather than peripheral.

What Schachter and Singer got right, broadly: emotion is not just physiology. Context matters. Cognitive interpretation matters. The phenomenology of an emotion is not a direct readout of bodily state.

What Schachter and Singer got wrong, specifically: the idea that arousal is a generic substrate, and that emotion is what you call it, is not supported by the empirical record. Emotions appear to be more differentiated at the physiological and neural level than the theory allowed, and the labeling mechanism is much more constrained than the experiments suggested.

How The Suspension Bridge Study (Dutton & Aron 1974) Fits In

The other piece of canonical evidence for the two-factor view in popular treatments is Donald Dutton and Arthur Aron’s 1974 “suspension bridge” study, often cited as the most vivid demonstration of “misattribution of arousal.”

In that study, male subjects crossed either a high, scary suspension bridge over a deep canyon, or a low, stable wooden bridge, and were then approached by an attractive female experimenter who asked them to fill out a short survey and offered her phone number “if they wanted to discuss the study further.” Men on the scary bridge called her back at higher rates than men on the stable bridge. Dutton and Aron interpreted this as misattribution of fear-based arousal as sexual attraction — the same theoretical commitment as Schachter and Singer, but in a field setting. (Dutton & Aron, 1974, DOI 10.1037/h0037031)

The suspension bridge study has its own replication issues. The original sample was small. There are confounds — men crossing a scary bridge may be different from men crossing a safe bridge in their baseline risk tolerance and willingness to cold-call attractive strangers. Several attempted replications have produced inconsistent results. Modern reviews of misattribution-of-arousal experiments find the effect is real in some narrow conditions (high but ambiguous arousal, attractive but plausible target), but is far less general than the popular framing implies.

The suspension bridge result is genuinely one of the prediction Reisenzein’s 1983 review concluded had some empirical support — the “intensification through misattribution” prediction, not the “labeling generates the emotion” prediction. So it is in the category of two-factor claims that partially survived. But it is also, as a single dramatic field study with a small sample and inconsistent replication, not robust evidence for the strong textbook version.

If the suspension bridge result is the strongest field evidence for two-factor theory and it is itself shaky, the lesson is the same as the lab evidence: the canonical story is shakier than the textbook treats it.

What’s Honest To Say About Emotion Theory Now

Here is a careful summary of what someone reading the empirical literature in 2026 should believe about emotion:

  1. Cognition shapes emotional experience. This is true and well-supported. How you interpret a situation affects what you feel about it.
  2. Context affects emotional experience. Also true and well-supported. The same physical event in different contexts produces different emotions.
  3. The body’s state contributes to emotional experience. True, but less straightforwardly than two-factor implies. Different emotions involve different physiological and neural patterns, not a single substrate that gets labeled.
  4. Emotion is not a direct readout of body state. True. There is no one-to-one mapping from physiology to felt emotion.
  5. Arousal is a generic substrate that gets labeled by context into specific emotions. This is the strict two-factor claim, and it is not well-supported. Arousal is more differentiated than that, labeling is more constrained than that, and unexplained arousal in the lab does not get neutrally redirected by social cues.
  6. The Schachter and Singer 1962 experiment, run as described, will produce the predicted pattern of emotional contagion. This is what the textbooks imply, and it is not supported by the 1979 replications or the subsequent literature.

A great deal of “emotional science” content circulating in business and marketing audiences — and a fair amount of “emotional intelligence” coaching content — assumes the strong two-factor version is settled science. It is not. The strong version failed replication. The attenuated version that survives is far less dramatic and far more conditional.

What This Means For Strategists Working With “Emotional Intelligence” Or “Emotional Arousal” Frameworks

If you are a marketing strategist, sales coach, conversion-rate consultant, or leadership development practitioner who has been quoting Schachter and Singer (or the suspension bridge study) as the foundational evidence for a framework about emotional arousal, here is the calibration you need to apply.

Claims that depend on the strong two-factor mechanism are on weak ground. Examples: “Get the customer aroused — excited, anxious, anything — and the brand will become the label for that arousal.” “Create high-stakes urgency and the buyer will misattribute the physiological state to genuine product enthusiasm.” “Make the demo emotionally intense in any direction and the prospect will translate that into wanting the product.” These claims sound scientific because they invoke a famous experiment, but the experiment they invoke did not replicate, and the strong form of the mechanism they assume is not supported by the evidence.

Claims that depend on the weaker, attenuated version are on firmer ground but explain less. Examples: “If a customer already has positive feelings about your brand, additional arousal can intensify those positive feelings.” “An emotionally engaging demo can amplify an existing inclination toward your product, but will not manufacture an inclination that wasn’t there.” These are reasonable applications of the surviving evidence — arousal intensifies existing emotional reactions through misattribution — and they make weaker, more conditional predictions than the popular framing.

General “emotional intelligence” frameworks are not in trouble. Many of the practical observations behind EI coaching — that recognizing your own emotional state, naming it, and considering its source improves decision-making — are reasonable, are supported by other lines of evidence (mood-as-information, affective forecasting, emotion regulation research), and do not depend on the two-factor theory specifically. You can keep coaching emotional self-awareness without relying on Schachter and Singer at all.

The right move for a strategist is to update the underlying theory while keeping the practical advice that has independent support. Stop citing the epinephrine experiment as the foundational evidence. Cite the modern emotion research — Barrett’s constructed emotion, Russell’s circumplex, the affective neuroscience of distinct emotional circuits. These give a more accurate picture and they support more careful, more conditional claims. The practical recommendations get more honest, not less useful.

The deeper lesson is the one this hub keeps coming back to: when a single dramatic experiment becomes the foundation for a wide-ranging applied framework, and the experiment turns out not to replicate, the framework needs to be rebuilt on better evidence — not defended on the strength of the original story. Schachter and Singer 1962 is one of the more important examples in social psychology because the theory it generated really did shape how generations of practitioners think about emotion, and the empirical foundation under that theory has been shaky for forty years.

Sources

  • Schachter, S., & Singer, J. E. (1962). Cognitive, social, and physiological determinants of emotional state. Psychological Review, 69(5), 379-399. DOI 10.1037/h0046234
  • Marshall, G. D., & Zimbardo, P. G. (1979). Affective consequences of inadequately explained physiological arousal. Journal of Personality and Social Psychology, 37(6), 970-988. DOI 10.1037/0022-3514.37.6.970
  • Maslach, C. (1979). Negative emotional biasing of unexplained arousal. Journal of Personality and Social Psychology, 37(6), 953-969. DOI 10.1037/0022-3514.37.6.953
  • Reisenzein, R. (1983). The Schachter theory of emotion: Two decades later. Psychological Bulletin, 94(2), 239-264. DOI 10.1037/0033-2909.94.2.239
  • Dutton, D. G., & Aron, A. P. (1974). Some evidence for heightened sexual attraction under conditions of high anxiety. Journal of Personality and Social Psychology, 30(4), 510-517. DOI 10.1037/h0037031
  • Cotton, J. L. (1981). A review of research on Schachter’s theory of emotion and the misattribution of arousal. European Journal of Social Psychology, 11(4), 365-397. DOI 10.1002/ejsp.2420110402
  • Carver, C. S., & Harmon-Jones, E. (2009). Anger is an approach-related affect: Evidence and implications. Psychological Bulletin, 135(2), 183-204. DOI 10.1037/a0013965
  • Barrett, L. F. (2017). The theory of constructed emotion: an active inference account of interoception and categorization. Social Cognitive and Affective Neuroscience, 12(1), 1-23. DOI 10.1093/scan/nsw154

FAQ

What about the suspension bridge effect — that’s the classic example everyone teaches?

The Dutton & Aron 1974 study is in the category of two-factor claims that has partial empirical support — the “arousal intensifies an existing emotional reaction” version, which Reisenzein’s 1983 review concluded was the one prediction with reasonable evidence. But the original suspension bridge study itself had a small sample and inconsistent replication, and modern reviews of misattribution-of-arousal experiments find the effect is real only in narrow conditions. It is not strong general evidence for the strict two-factor mechanism.

Is the textbook version of Schachter and Singer really wrong?

The textbook version usually says: same arousal, different context, different emotion — neatly demonstrated by the epinephrine experiment. The actual empirical record is: the original 1962 study had methodological issues that were apparent at the time, two careful 1979 replications failed to reproduce the central findings, and the comprehensive 1983 review concluded that the strong form of the theory was not supported. So yes — the textbook version is misleading. The version that survives is much more conditional and explains less.

What about Lisa Feldman Barrett’s theory of constructed emotion?

Barrett’s theory of constructed emotion is the leading modern alternative to discrete-emotion theories and to strict two-factor theory. In her view, emotions are concepts the brain constructs from interoceptive predictions, sensory input, and learned categories. This is a substantially more sophisticated framework than two-factor theory and is supported by a much larger evidence base in cognitive neuroscience. It preserves the spirit of “cognition matters for emotion” while abandoning the specific mechanism Schachter and Singer proposed.

Should I stop using “emotional intelligence” frameworks in coaching or leadership work?

No, but you should not rely on Schachter and Singer as the foundational science for them. Most of the practical advice in EI coaching — recognize emotions, name them, consider their source, regulate responses — has independent support from other lines of evidence and does not depend on the two-factor theory. Update your underlying citations to modern emotion research and the practical recommendations get more accurate, not less useful.

If the theory failed, why is it still in every introductory textbook?

Textbooks are extremely slow to update. The Schachter and Singer story is vivid, memorable, easy to teach, and fits a clean narrative arc (debate between James-Lange and Cannon-Bard, resolved by a clever experiment). Pulling it would require restructuring the emotion chapter around theories that are harder to summarize in three paragraphs. The 1979 replication failures and the 1983 review have been in the literature for over forty years; the textbook treatment has not caught up. This is one of many examples in psychology of textbook content lagging the actual empirical state of the field by decades.

Does this affect claims in sales and marketing that “getting the customer emotionally aroused” causes them to attribute the arousal to product enthusiasm?

Yes — those claims are on weaker ground than they sound. They are usually implicitly invoking the strong two-factor mechanism (undifferentiated arousal gets labeled by context), which is the part of the theory that did not replicate. The narrower, surviving version of the mechanism (arousal can intensify an existing emotional reaction) makes much weaker predictions. Marketing strategies that depend on manufacturing arousal-then-relabeling are unlikely to work as advertised. Marketing strategies that depend on amplifying an emotional response the customer already has are on firmer ground but also describe a less dramatic effect.

Is there any version of two-factor theory that is empirically supported in 2026?

The attenuated version Reisenzein identified in 1983 still seems roughly right: arousal feedback from one source can intensify an emotional reaction with a different cause, when the cause of the arousal is ambiguous and the attribution is plausible. That is a real mechanism with real empirical support. But it is a much smaller claim than the textbook two-factor theory, and it does not justify the broader “emotion is a label on undifferentiated arousal” framework that the original paper inspired.

What should I cite instead if I want to talk about how cognition shapes emotion?

Lisa Feldman Barrett’s work on constructed emotion is the leading modern theoretical framework. James Russell’s circumplex model of affect is a useful framework for valence-arousal space. Carver and Harmon-Jones (2009) is good for distinguishing approach- versus avoidance-related emotions. The affective neuroscience literature on distinct emotional circuits (e.g., Panksepp’s seven primary emotional systems) is good for the biology. Any of these is a stronger foundation than two-factor theory for an applied framework about emotion.

replication-crisis schachter-singer emotion-theory social-psychology evidence-evaluation
Share this article
LinkedIn (opens in new tab) X / Twitter (opens in new tab)
Atticus Li

Experimentation and growth leader. CXL-certified CRO practitioner, Mindworx-certified behavioral economist (1 of ~1,000 worldwide). 200+ A/B tests across energy, SaaS, fintech, e-commerce, and marketplace verticals.

About LinkedIn Newsletter