BACKGROUND: Decision analytic models typically compare strategies using expected values, which assumes risk neutrality in all situations for all individuals. OBJECTIVE: Examine how risk attitude is affected by the magnitude of the risk involved, and by the sociodemographic characteristics of the patient. METHODS: A set of 5 related standard gambles was created. In all of them the patient has an asymptomatic disease that will cause painless death in 5 years. The only treatment is a pill that only works if taken immediately. If successful, the patient will live for 15 years. The 5 scenarios correspond to 0-4 years until death in the event of treatment failure. The probability of failure was varied to establish the break-even point. The expected break-even point was calculated based on expected values. Risk attitude was measured using the Risk Attitude Ratio (RAR), the ratio of the individual's break-even point to the expected break-even point. RAR=1 indicates perfect risk neutrality, with lower values indicating risk adversity. Regression models for each scenario were constructed for RAR, with regressors based on age, race, sex, education, income and marital status. Additionally, a hierarchical linear model was constructed using data from all 5 scenarios. RESULTS: 167 people were surveyed. Significant risk adversity was observed in all scenarios. Mean break-even points ranged from .07-.35. When treatment failure meant immediate death, risk aversity was great with no statistically significant associations. As time until death increased, RAR increased (less risk averse) and was associated with sex, race and marital status. These associations were upheld in the overall model. Males and the previously married were less risk averse. African-Americans were more risk averse. CONCLUSIONS: Risk attitude varies according to the specifics of the gamble as well as socio-demographics. Assumptions of risk neutrality may be particularly poor in cases where treatment failure carries a severe penalty.