Subjects evaluated 27 hypothetical job offers using both riskless rating scale and risky utility response modes. These preference judgments were then analyzed using both statistical and conjoint measurement procedures. Additive statistical models provided excellent approximations to both the riskless and risky responses of every subject. Nonetheless, half of the subjects showed statistically significant, and in some cases, substantial departures from additivity. Despite the predictive success of the additive and multiplicative models studied, the conjoint measurement analysis revealed that several subjects substantially violated the ordinal properties shared by both additive and multiplicative models. In addition, all subjects showed some minor violations of the conjoint measurement axioms. A simple numerical analysis showed, however, that most of these minor violations may have been due to random error. © 1976.