A theoretical model of the relationship between user expectations and the changing state of a User Adapted Interface is presented. The usability of an intelligent email client that learns to filter spam emails is tested under three variants of adaptation: no user modeling, user modeling with fixed (optimal) spam cut-offs, and user modeling with user adjustable spam cut-offs. The results supported our hypothesis that user control over adaptation is preferred over no control because the user can maintain the system?s interaction state within their region of expectation. The study suggests that this remains true even when performance of the system (accuracy of spam filtering) degrades because of errors in user control (adjustment of spam cut-offs).