A program of research began with the goal of creating a flexible, computerized adaptive assessment of personality characteristics that was resistant to faking and predictive of important aspects of job performance. The latent structure of personality was explored to identify facets underlying the Big Five that were sufficiently unidimensional for item response theory (IRT) to be utilized. We found that an ideal point item response theory model was needed to adequately describe responding to personality questions; more commonly used dominance models such as the two-parameter logistic were found to fit poorly. A two-alternative forced choice format has proven fake resistant, provided that statements are balanced on both social desirability and dimension extremity. With a computer adaptive algorithm, test length can be half the length of a static assessment without loss of reliability. Predictive validity results for large samples of U.S. Army soldiers will be presented.