Much of the empirical data available about therapeutic alliance and its relationship to termination status come from individual psychotherapies. We know less about therapeutic alliance in couple therapy. A unique characteristic of alliance in couple or family therapy is the possibility of discrepancies in alliance between system members. In this study we sought to demonstrate three statistical techniques: standard deviations, the intraclass correlation to assess discrepancies in alliance over time during the initial stage of couple therapy, and the use of these various measures to predict termination status using a sample of 72 couples from a university-based training clinic. Differences in partners’ alliances operationalized either as categorical or continuous variables but when analyzed separately at each time point were not predictive of termination status. When multilevel modeling was used, a difference in the way the discrepancies changed over a period of time was related to termination status. Rashmi Gangamma is currently at Syracuse University.