Three methods of item selection were used to select sets of 38 items from a 50-item verbal analogies test and the resulting item sets were compared for internal consistency, standard errors of measurement, item difficulty, biserial item-test correlations, and relative efficiency. Three groups of 1,500 cases each were used for item selection. First it was determined that, using a limit of 10 for mean square fit, there were 38 items that fit the Rasch model. Then the 38 items with the highest biserial item-test correlations were selected from the 50-item test, and the 38 items having the highest loadings on the first unrotated principal component were selected to make up the three item sets for comparison. There were 29 items common to all three sets. The 4,500 cases used for item selection were then randomly reassigned to three new samples to evaluate the three tests. The items selected by the Rasch method resulted in the most difficult and least internally consistent test, but all differences were small. Lord's procedures for approximating relative efficiency also showed the three tests to be quite similar over most of the score range. (CTM)