Learning to Learn (LTL) is the transfer of learning, separate from stimulus generalization, that appears across tasks that share a similar structure. Three experiments examined this phenomenon in both conditioning and extinction learning in humans. The latter effect is of special interest given the failures in the literature to obtain transfer of extinction between stimuli. Conditioning and extinction with one stimulus increased the rate of conditioning and, surprisingly, extinction of a different stimulus (Experiment 1). The effects appeared in the absence of physical generalization. The transfer of extinction was not enhanced by conditions that increased the chances of a mediated extinction effect (Experiment 2). Finally, Experiment 3 ruled out three possible sources for the effect in extinction: a common unconditioned-stimulus representation, a common reinforcement history, and within-stimuli associations. Overall, the findings are consistent with the idea that LTL is an emergent (non-immediate) form of mediated generalization that is dependent upon memory structures retrieved by trial outcomes. The over- or under-prediction of the outcome on the first trial with a new task might retrieve prior episodes associated with similar prediction errors promoting transfer.