Today, Internet and Web technologies not only provide students opportunities for flexible interactivity with study materials, peers and instructors, but also generate large amounts of usage data that can be processed and reveal behavioral patterns of study and learning. This study analyzed data extracted from a Moodle-based blended learning course, to build a student model that predicts course performance. CART decision tree algorithm was used to classify students and predict those at risk, based on the impact of four online activities: message exchanging, group wiki content creation, course files opening and online quiz taking. The overall percentage of correct classifications was about 99.1%, proving the model sensitive to identify very specific groups at risk.