Predictive solubility models can be of great use for crystallization modeling and optimization, and can decrease the amount of experimental effort needed to create a robust crystallization model. In this study, predictive solubility models such as MOSCED, UNIFAC, NRTL-SAC, and the Jouyban-Acree model are compared against an empirical model for predicted solubility accuracy. The best models are subsequently compared against the empirical model for the antisolvent crystallization of acetaminophen in acetone, using water as the antisolvent. The effects of these solubility models on the predicted relative supersaturation, volume mean size, volume-percent crystal size distribution (CSD), and generated optimal antisolvent feed profiles are investigated. It was found that, for this system, only the NRTL-SAC and Jouyban-Acree solubility models were accurate enough to predict crystallization mean size and crystal size distributions. The Jouyban-Acree and NRTL-SAC solubility models respectively predicted end-volume mean-size differences up to 13% and 29% from the empirical model. When used to create optimal antisolvent feed profiles, the Jouyban-Acree and NRTL-SAC profiles produced results that varied up to 32% and 60%, respectively, from the desired objective. None of the predictive solubility models was accurate enough for the creation of optimal antisolvent feed profiles.