A study examined the predictive validity of two methods of prior knowledge assessment: free association and prediction. Subjects, 72 sixth graders from suburban public and private schools, were divided into four groups: standard free association; standard free association plus prediction; definitional free association; and definitional free association plus prediction. Subjects read two expository and two narrative texts, were asked to free associate and/or predict passage elements, retold the passage, and answered comprehension questions. Data from the 57 students whose comprehension was above 50% on at least one passage were subjected to multivariate analysis. Results indicated that: (1) although definitional instructions did not yield higher prior knowledge scores than free association instructions, the correlations between the knowledge scores and comprehension of narrative text for the two instructional groups were different; (2) in expository text, none of the correlations between responses to concepts and retelling or comprehension were significant in either instructional condition; (3) prediction of main ideas prior to reading was correlated with retelling and comprehension of expository text, but less so with narrative text; and (4) the number of correct main ideas predicted was very low. Findings suggest that definitional instruction tapped more directly what students knew, and did not know, about concepts. (Nineteen references are attached.) (RS)