This paper analyses the empirical risk tolerance of individuals and the role of physiological measures of risk perception. By using a test that mimics the financial decision process in a laboratory setting (N = 445), we obtained an ex-post empirical measure of individual risk tolerance. Predictive classification models allow us to evaluate the accuracy of two alternative risk-tolerance forecasting methods: a self-report questionnaire and a psycho-physiological experiment. We find that accuracy of self-assessments is low and that misclassifications resulting from questionnaires vary from 36 to 65%: individuals asked to self-evaluate their risk tolerance reveal a high probability of failing their judgement, i.e. they behave as risk takers, even if, before the task, they define themselves as risk averse (and vice versa). Conversely, when the risk-tolerance forecast is obtained from individuals’ physiological arousal, observed via their somatic activation before risky choices, the rate of misclassification is...