Knowing what predicts discontinuation or success of psychotherapies for Borderline Personality Disorder (BPD) is important to improve treatments. Many variables have been reported in the literature, but replication is needed and investigating what therapy process underlies the findings is necessary to understand why variables predict outcome. Using data of an RCT comparing Schema Therapy and Transference Focused Psychotherapy as treatments for BPD, variables derived from the literature were tested as predictors of discontinuation and treatment success. Participants were 86 adult outpatients (80 women, mean age 30.5 years) with a primary diagnosis of BPD who had on average received 3 previous treatment modalities. First, single predictors were tested with logistic regression, controlling for treatment type (and medication use in case of treatment success). Next, with multivariate backward logistic regression essential predictors were detected. Baseline hostility and childhood physical abuse predicted treatment discontinuation. Baseline subjective burden of dissociation predicted a smaller chance of recovery. A second study demonstrated that in-session dissociation, assessed from session audiotapes, mediated the observed effects of baseline dissociation on recovery, indicating that dissociation during sessions interferes with treatment effectiveness. The results suggest that specifically addressing high hostility, childhood abuse, and in-session dissociation might reduce dropout and lack of effectiveness of treatment.