Animals are sometimes sensitive and sometimes indifferent to variability in amount of reward. I analyzed a large set of published studies to test whether these differences in risk-sensitivity may be influenced by the coefficient of variation (s.d./mean) of the variable reward and by the presence of empty rewards. The coefficient of variation had a strong effect on the strength of risk-sensitivity among studies with nectarivores, both invertebrates and vertebrates. The effect was not significant for studies with non-nectarivores, but these studies included only a limited range of coefficients of variation, and risk-sensitivity in these studies was based on fewer subjects per experimental condition than in studies of nectarivores. The presence of empty rewards had no effect on the strength of risk-sensitivity. The apparent effect of empty rewards seems to be an artifact of the tendency of distributions that include zero rewards to have larger coefficients of variation than distributions that do not include zero rewards. I introduce the CV model, a simple descriptive model that accounts for a large proportion of the variability in levels of risk-sensitivity among experiments. My analysis resolves the apparent discrepancies among studies that report risk-sensitivity and risk-indifference. It is necessary to control for the magnitude of the coefficient of variation when testing risk-sensitivity under different conditions (e.g., energy budgets). A successful descriptive model of risk-sensitive choice behavior is useful for guiding future research both in ecology and in the underlying mechanism of cognitive processes.