The automatic monitoring of human activities, with the purpose of determining the level of autonomy of an elderly person, is a challenging research field in Ambient Intelligence. Existing solutions follow one of two different approaches:s mart environments rely on heterogeneous sensors distributed in the environment; wearable sensing systems rely on sensors located on the person body. The article proposes a framework for the recognition of simple human activities that requires limited computational resources and that is based on the information coming from a single wrist-placed tri-axial accelerometer. The system relies on Gaussian Mixture Modelling and Gaussian Mixture Regression for the creation of models of the activities and on Mahalanobis distance for the classification of the run-time data.