Food consumers often query or ignore the risk assessments of scientists, the food industry and public bodies. This is widely acknowledged. It has been suggested that this ‘expert-lay discrepancy’ is a relatively straightforward upshot of the fact that lay people lack the knowledge and technical understanding of experts. However, much published research on risk in psychology and sociology runs counter to this ‘knowledge deficit’ model (Hansen et al., 2003). In many cases, at least, lay risk assessments are not well explained as the product of ignorance, because they are in fact complex, situational sensitive expressions of a person's value system. There is obviously a pressing need today to understand expert-lay discrepancies in the assessment of food risks. We need to know how consumers perceive and assess risks; why they respond to communication and advice relating to those risks in the way they do; what factors affect their willingness to trust public institutions responsible for regulating the food industry and issuing guidance on food matters; and what determines their handling of specific food hazards. Psychometric and psychological studies of risk perception offer an invaluable corrective to excessive and simplistic reliance on the deficit model (Fife et al., 2000; Hansen et al., 2003). By emphasising the multi-dimensionality of lay risk perception, they have improved the understanding of expert-lay discrepancy. More generally, the demonstration that risks and benefits are not perceived independently of each other is a crucial finding of psychological research, and we now know that risk-benefit analyses that treat risk and benefits as independent factors should be handled with caution by those aiming to interpret or influence popular opinion. Our paper is an attempt to analyse consumers’ behaviour when facing a potential risky action such as consuming GM food. The hypothesis tested states that consumers take into account both costs (accident) and benefits (rewards) of uncertain outcomes and then minimize risk instead of trying to totally avoid it. A sample of 338 students, interviewed in year 2000 (188) and in year 2008 (150), enrolled in different Italian colleges was directly interviewed on an hypothetical genetically modified tomato market. A mixture distributions was used first for inferring on what variables influence the decision to take part on the "new market" proposed and, second, to estimate the Willingness To Pay (WTP) distribution for those willing to buy the GM product proposed.