During the past decade, personalized e-learning systems and adaptive educational hypermedia systems have attracted much attention from researchers in the fields of computer science Aand education. The integration of learning styles into an intelligent system is a possible solution to the problems of "learning deviation" and "cognitive overload." In this study, we propose a learning style prediction method based on a pattern recognition technique. The main contributions of this method are: (1) it is a form of middleware that can be applied to other intelligent tutoring systems, and (2) it can process topic-dependent data to make predictions and update learning style profiles in a recursive manner. Experimental evaluations demonstrated the effectiveness of this prediction method.