This paper presents an ACT-R model of a realistic menu item selection task from a previous study. The model aims to explain observed trends in both the mean and dispersion of response time (RT) with respect to both menu length and amount of practice. Our model follows previous work in assuming that the primary strategies for determining the location of the target menu item are visual search and recall from memory. The model introduces a hypothesis about how the visual search is performed that yields a very good fit with no free parameters for the slope of the linear relationship between RT and menu length. We validated this hypothesis by testing the model's predictions for the dispersion of RT. The model was also used to generate predictions to test the hypothesis that the observed RT reduction with practice could be predicted based on RT being the sum of the time required for initiating the task and the lesser of the realized visual search time or realized recall time. The model generated predictions for both means and dispersion of RT as a function of practice.