Given observed stochastic choice data and a model of stochastic choice, we offer a methodology that enables separation of the data representing the model's inherent randomness from residual noise, and thus quantify the maximal fraction of the data that are consistent with the model. We show how to apply our approach to any model of stochastic choice. We then study the case of four well-known models, each capturing a different notion of randomness. We conclude by illustrating our results with an experimental dataset.