We show that observed choices in discounting experiments are consistent with roughly one-half of the subjects using exponential discounting and one-half using quasi-hyperbolic discounting. We characterize the latent data generating process using a mixture model which allows different subjects to behave consistently with each model. Our results have substantive implications for the assumptions made about discounting behavior, and also have significant methodological implications for the manner in which we evaluate alternative models when there may be complementary data generating processes.