This paper describes eight different strategies that subjects might use to allocate their short-term memory resources during reading. Two experiments are then presented to determine which of these provides the best description of human performance. In the first experiment the eight strategies were used to predict the contents of readers' verbal protocols. In the second experiment they were used to predict the readability of 20 texts. In both experiments, a strategy based on the plans and goals of characters in the texts provided the best fit between theory and data.