In order to better examine seemingly unpredictable variation that appears in decision-making studies, I had people chose between two options that had no features or consequences to distinguish them. 100 users of Mechanical Turk completed 200 binary choices, and I examined the accuracy with which statistical models could predict the choices. Across three different conceptualizations of the prediction problem and a variety of models ranging from logistic regression to neural networks, I obtained at best modest predictive accuracy. Predicting trivial choices may actually be more difficult than predicting meaningful choices. These strongly negative results appear to place limits on the predictability of human behavior.