We propose a new personalized recommendation technique, which dynamically recommends products based on user behavior patterns for E-commerce. It collects and analyzes user behavior patterns from XML-based E-commerce sites using the PRML (Personalized Recommendation Markup Language) approach. The collected information is saved as PRML instances and an individual user profile is built from the PRML instances of the user using a CBR (Case-Based Reasoning) learning technique. When a new product is introduced, the system compares, for a user, the preference information saved in the user profile and the information about the new product and produces a recommendation that best fits the user preference.