The monitoring of human activities plays an important role in health-care applications and for the data mining community. Existing approaches work on activities recognition occurring in sensor data streams. However, regular behaviors have not been studied. Thus, we here introduce a new approach to discover top-k most regular episodes from sensors streams, TKRES. The top-k approach allows us to control the size of the output, thus preventing overwhelming result analysis for the supervisor. TKRES is based on the use of a simple top-k list and a k-tree structure for maintaining the top-k episodes and their occurrence information. We also investigate and report the performances of TKRES on two real-life smart home datasets.