College students made choices in binary lottery situations involving small amounts of money-payoffs ranged from $2.50 to -$1.00. Each of 52 subjects made 150 choices. Any of a number of choice models might be used to describe the outcome of such an experiment. However, no simple model (based on expected utility maximization) is capable of explaining all choices (of a particular subject)-there are always some inconsistent choices. The experiment was designed to isolate factors affecting inconsistency. Such research is aimed at improving the predictive ability of choice models by reducing the stochastic error. A choice model based on expected utility maximization was developed. The model was designed to maximize the internal consistency of each subject's choices. In all cases, some choices were found to be inconsistent with the model. Grouping the data in various ways, four factors hypothesized to affect inconsistency were tested: difference in dispersion of the payoffs, signs of the payoffs, order of the choice situation, and small changes in wealth. Applying classical statistical techniques, inconsistency was found to be related to the first three of these factors. Some implications of these results for expected utility models of choice are presented.