Objectives. This study hypothesized that several baseline client characteristics (i.e. age, symptoms, insight, social functioning) would significantly predict client-rated group alliance in out-patients with schizophrenia spectrum disorders. Design. Hierarchical linear modeling (HLM) was used to evaluate the contributions of selected baseline individual client characteristics and group level characteristics to client-rated group alliance at the sixth session of group therapy. The effect of treatment type (CBT vs. ST) on group alliance and interaction with predictor variables were also analysed. Finally, correlations were computed to explore the relationship between group alliance, attendance, and treatment engagement. Methods. Sixty-three out-patients who had treatment-resistant auditory hallucinations were randomly assigned to either group CBT, which targeted reduction of distress associated with hallucinations, or group ST, which focused on improving social integration. Results. Results indicate that a stronger group alliance at the mid-point of treatment was associated with overall higher levels of group insight, and lower individual autistic preoccupation and social functioning at the baseline assessment. In addition, stronger group alliance was significantly correlated with higher attendance rates and therapists' ratings of treatment compliance. Conclusions. These findings have implications for determining group composition and identifying clients low in therapeutic engagement. Suggestions for future research on group alliance are also discussed.