When the three-parameter logistic model and item response theory are used to analyze Graduate Management Admission Test (GMAT) data, there are problems with the assumption of unidimensionality. Linear factor analytic models, exploratory factor analysis programs, and the comparison of item parameter estimates for heterogeneous and homogeneous subsets also present difficulties. A new method of assessing the dimensionality of binary data is now available. TESTFACT is a computer program which can be used to perform full information factor analysis, using the marginal maximum likelihood method to estimate reparameterized discrimination and difficulty parameters for multidimensional item response models. The lower asymptote for each item is treated as a known constant whose value is input by the program user. TESTFACT allows a stepwise factor analysis to be performed. First a one-factor solution is performed, then a two-factor solution. The difference between chi squares for the two solutions is used to test whether the added factor is statistically significant. When TESTFACT was applied to both quantitative and verbal GMAT items, a dominant first factor and two additional, considerably weaker, factors were found. (Author/GDC)