Ramsay curve item response theory (RC-IRT) was recently developed to detect and correct for nonnormal latent variables when unidimensional IRT models are fitted to data using maximum marginal likelihood estimation. The purpose of this research is to evaluate the performance of RC-IRT for Likert-type item responses with varying test lengths, sample sizes, and latent variable distributions. Also, RC-IRT is compared to a related procedure implemented in the PARSCALE program. Results indicated that for nonnormal latent variables, item parameters and scores from RC-IRT can be more accurate than those obtained from the normal model, as long as there is enough information in the data (i.e., enough items and people). The procedure in PARSCALE tended to give better answers than the normal model but was generally not as accurate as RC-IRT. (Contains 6 tables, 3 figures and 2 notes.)