Students using a Learning Management System (LMS) as a learning support have been observed to demonstrate different learning behaviours. Studies have reported students exhibiting different procrastination tendencies, distinct social behaviours and system usage patterns. Students can be clustered together based on similarity in their learning behaviours. The K-means clustering algorithm is a simple and effective way to group students with similar behaviours. The authors use this algorithm in a novel way. It is applied in two phases on an unlabelled dataset obtained from LMS course logs. In the first phase, distinct clusters are formed using K-means. In the second phase, K-means is again applied to clusters obtained in the first phase to obtain further insight into students' interaction behaviours. The two-phase application of K-means clustering clearly revealed the most and least successful learning behaviours. The authors also establish a relationship between observed behaviours and course final scores.