We study time preferences by means of a longitudinal lab experiment involving both monetary and non-monetary rewards (leisure). Our novel design allows to measure whether participants prefer to anticipate or delay gratification, without imposing any structural assumption on the instantaneous utility, intertemporal utility or the discounting functions. We find that most people prefer to anticipate monetary rewards (positive time preferences for money), but they delay non-monetary rewards (negative time preferences for leisure). These results cannot be explained by personal timetables and heterogeneous preferences only. They invite to reconsider the psychological interpretation of the discount factor, and suggest that the assumption that discounting is consistent across domains can lead to non-negligible prediction errors in models involving non-monetary decisions, such as labor supply models.