Three test-analysis models were used to analyze three types of simulated test score data plus the results of eight achievement tests. Chi-square goodness-of-fit statistics were used to evaluate the appropriateness of the models to the four kinds of data. Data were generated to simulate the responses of 1,000 students to 36 pseudo-items by procedures assuming a three-parameter, a two-parameter, and a one-parameter model. These three sets of pseudo-responses plus the data from eight achievement tests given to seventh and eighth grade children were analyzed by the three methods. The three-parameter logistic model provided the best fit in most instances. Ability estimates based on the two-parameter and the three-parameter models had a curvilinear relationship. The two-parameter estimation of item discriminating power was unacceptable, being strongly affected by the item's difficulty level. Data for the eight actual multiple-choice tests were most consistent with the three-parameter model. (CTM)