Measurement disturbances, such as guessing, startup, and plodding, often result in an examinee's ability being either over- or under-estimated by the maximum likelihood estimation employed in latent trait psychometric models. Several authors have suggested methods to lessen the impact of unexpected responses on the ability estimation process. This study uses simulated data to compare the standard maximum likelihood estimation with two forms of robust estimation, BIWEIGHT and AMJACK, and person analysis within the Rasch model. The results indicate that, although the two robust estimation procedures recover the generating parameters under certain conditions, in the presence of many forms of measurement disturbances they mask important information. Rasch person analysis has the advantage of not only providing a method of modifying the ability estimation procedure but also of providing a means of identifying the nature of the disturbance. (Author)