Cohesiveness has been viewed as the group psychotherapy equivalent of the therapeutic alliance in individual treatment. Although researchers have attempted to study the concept of cohesion in group treatment, understanding of this so-called "curative" group factor remains quite primitive. In this study of 12 time-limited psychotherapy groups, with a total of 90 nonpsychotic outpatients, we explore the relationships between cohesion, alliance and treatment outcome. Our cohesion measure is a new instrument, the Harvard Community Health Plan Group Cohesiveness Scale, developed for use with group therapy videotapes. To measure alliance we have modified the Penn Helping Alliance Scale (Group Alliance Scale) to be scored from videotapes of group sessions. Both of these instruments use trained observers to make ratings for the group-as-a-whole. The outcome battery for patients in these groups included a widely varied set of measures, enabling us to view change from a number of perspectives. Our findings indicate that cohesion and alliance as measured here are related concepts. We also find that both cohesion and alliance appear to have strong relationships with improved self-esteem and reduced symptomatology for patients in these groups. In addition, it appears that outcome is most related to cohesion in the first 30 minutes of a group session. Implications of these and other findings are discussed.