The focus of this work is to explore the possibilities of recognizing three common user activities (sitting, walking and running) with accelerometer data from smartphones. Among five common machine learning algorithms, Naïve Bayes classifier proved to be the best choice. Classification accuracy of more than 90% was achieved when phone is carried in a pocket. It is shown that this method is appropriate and that the phone’s orientation information is not needed. Finally, the classification of one day-long data set is presented.