We investigated how people understand and recall simple stories. After discussing our general framework for investigating memory, we examined story grammars considered as theories of readers’ memory of a story. Story grammars were found to be inadequate as grammars, as recognition devices for stories, and as predictors of recall probabilities of different statements in three test stories. An alternative approach views a story as a problem-solving protocol and analyzes it into a hierarchical state transition (HST) network; actions were viewed as succeeding or failing to bring about state changes, with actions perhaps being decomposed into subactions. We hypothesized that successful actions, and those higher in the action hierarchy, would be remembered titter. Recall evidence supported these hypotheses. First, they predicted recall of statements within our three test stories. Second, people recalled action sequences that were completed, or that succeeded in attaining a goal, better than ones that were begun but abandoned before completion or because they failed. Third, superordinate actions were recalled more than subordinate ones. Fourth, setting information that enabled plot actions was recalled more than unused setting information; moreover, used settings tended to be misrecalled near actions they enabled. Finally we discussed incompletenesses of the HST approach. It requires a processing theory. We suggested some of its components and some story phenomena it must encompass.