Discrete choice methods are often used for the estimation of time preferences. We show that these methods have pervasive problems when based on random utility models, for which cases our results establish that the probability of selecting a later option over an earlier one may be greater for higher levels of impatience. This could have profound implications, not only in the experimental estimation of time preferences, but also in a wide variety of empirical papers using such models in dynamic settings. Alternatively, we also show that discrete choice methods built on random preference models are always free of all such problems.