Résumé : In this paper we applied data fusion approaches for predicting the final academic performance of university students using multiple-source, multimodal data from blended learning environments. We collected and preprocessed data about first-year university students from different sources: theory classes, practical sessions, on-line Moodle sessions, and a final exam. Our objective was to discover which data fusion approach produced the best results using our data. We carried out experiments by applying four different data fusion approaches and six classification algorithms. The results showed that the best predictions were produced using ensembles and selecting the best attributes approach with discretized data. The best prediction models showed us that the level of attention in theory classes, scores in Moodle quizzes, and the level of activity in Moodle forums were the best set of attributes for predicting students' final performance in our courses.