Script actions that varied both in relevance to the script and in the expectancy of their details were developed. In Experiment 1, recall of these actions was tested. Actions of medium relevance were recalled better than actions of high or low relevance, whether or not details were presented. However, the unexpected details themselves were recalled better than the expected details. In Experiment 2, recognition was lowest for high-relevance actions but medium-and low-relevance actions did not differ. Actions with unexpected details were recognized better than actions with expected details, which tended to be better than generic actions. The results were interpreted as showing that both the relevance and the expectancy of the details in script actions are important in memory, but that the two dimensions may play different roles.Many studies have shown that prior knowledge affects the processing of new material-that schemata are used to comprehend and organize integrated material (Mandler, 1984;Taylor & Crocker, 1981). Generally, information important to an activated schema is remembered better than information less important (Johnson, 1970); however, other findings appear to conflict with this generalization. Using scripts, which are highly schematized sets of actions, Graesser, Woll, Kowalski, and Smith (1980) found that actions that were highly typical, and presumably important, were recalled and recognized more poorly than actions that were less typical. In person memory experiments, Hastie (1980) and Srull (1981;Srull, Lichtenstein, & Rothbart, 1985;Srull & Wyer, 1989) found that behaviors incongruent with trait expectations were recalled better than behaviors that were congruent. Thus, the literature on importance indicates that the more related material is to a schema, the better it will be remembered; but the script and person memory literatures seem to show the opposite.The purpose in the present experiments was to understand when relatedness to a schema helps memory and when it hurts memory. Relatedness was manipulated in two ways. First, relevance of actions to an activated script was varied with low-, medium-, and high-relevance actions. The main question for this variable was whether relevance would produce an inverted U-shaped function in recall and a negative linear function in recognition. Second, the expectancy of the details contained in those actions was varied. The main question here was whether both recall and recognition would be better for unexpected Appreciation is expressed to Kim Veum, Heidi Goos, Tom Tigue, and Nate Emerson for testing the subjects and scoring the data. Portions of this paper were presented at the annual meeting of the Psychonomic Society, seattle, 1987. Requests for reprints should be sent to