Interface evaluation is a necessary phase in the production of quality user interfaces. The usual evaluation techniques involve formal experiments or observation, and can be invasive. One non-invasive method that can be used at user sites is to record all user input and system output to a file. This transcript is then algorithmatically analyzed to determine interface problems. A new technique analyzes these transcripts by searching for maximal repeating patterns (MRPs), on the hypothesis that repeating sequences of user actions indicate interesting user behavior, and therefore may show problems in the interface. The technique was tested by using it to evaluate the human-computer interface of a large and complex image processing system in active use. Results show MRPs useful in detecting specific problems within the interface.