Four new methods of item analysis were used to select subsets of items which would yield measures of attitude change. The sample consisted of 263 students at Michigan State University who were tested on the Inventory of Beliefs as freshmen and retested on the same instrument as juniors. Item change scores and total change scores were computed for each subject. Responses of half the sample were used for item analyses. The four methods of change item analysis employed were: selection on the basis of high change item score variance; selection on the basis of pretest response frequency; selection on Saupe's correlation between change item score and total change score; and selection on triserial correlation between item change score and total change score. Subsets of 15, 30, 60, and 90 items were chosen by each method. In addition, subsets of equal size were randomly selected. When change score reliability was computed, using the responses of the cross-validation group, all four methods of item analysis resulted in higher change score reliability than did random selection. (Author)