Can a toolbox of simple heuristic rules help explain lottery choices relative to expected utility theory (EUT)? While a mixture model of EUT plus heuristic rules will obviously fit data better than EUT only, given the small sample sizes, there is a danger of overfitting. Therefore, instead of goodness-of-fit measures, we focus on forecasting performance. Using two data sets of binary lottery choices and reasonable holdout subsets for testing forecasting performance, we find that the EUT-only model forecasts better than the toolbox mixture model with EUT. Even when the toolbox model with EUT fits the data significantly better, EUT-only forecasts better.