The Working Alliance Inventory (WAI) is the most widely used measure of the therapeutic alliance. However, previous studies of the factor structure of the WAI have obtained disparate results. This study examined ratings from three rater perspectives (therapists, clients, and observers) in a rehabilitation program for high-risk violent prisoners with high PCL-psychopathy scores. The authors used confirmatory factor analysis with a short version of the WAI and examined one-, two-, and three-factor models. It was hypothesized that the context and client characteristics could alter the factor structure of the ratings. However, no differences were found in factor structure by perspective; the results from all rater perspectives suggested that a two-factor structure was the best fit for the data. Thus, we concluded that therapists, observers, and prisoners all saw the goals and task aspects of the therapeutic alliance as distinct from the bond component.