We present a laboratory study investigating the generalization of learning across two games of strategic interaction. The participants’ performance was higher when a game was played after, as compared to before, a different game. We found that the generalization of learning from one game to another was driven by both surface and deep similarities between the two games. We developed a computational cognitive model to investigate mechanisms of generalization. Model development highlighted some of the challenges of cognitive modeling in general and modeling strategic interaction in particular. We found that development of reciprocal trust was a key factor that explained the observed generalization effect.