A cross-validated study was performed to assess the ability of a "rule-finder" program to generate rules describing everyday behaviour. Sixty-four participants produced written reports about recent sequences of action in their everyday lives. These were input to the program as groups of numbers rperesenting certain features, such as the time of day, the other people involved, and the reason for the action. Ten rules emrged which could predict the type or features of an unknown action in a new sample of cases with significant accuracy. This way of describing behaviour has a number of advantages. The rules can also be converted fairly simply into a "production system" for use in further modelling work.