An increase in the scale of modern science is associated with the proliferation of a new kind of research formation: collaborations involving teams of researchers from several organizations. Historical and sociological studies indicate substantial variation in such formations, but no general classification scheme exists. The authors provide the outline of a scheme through a systematic analysis of multi-institutional collaborations that span a variety of fields in physical science. First, general dimensions of scientific collaborations were identified through a qualitative, historical analysis (...) of collaborations in high energy physics, space science, and geophysics. Next, the authors used informants in five new areas to collect systematic information on twenty-three recent collaborations. Finally, cluster analysis was employed to develop a variety of classification schemes and examine their relationships with outcome dimensions. Results show that a classification based on a broad conception of technological practice is superior to others in its ability to predict perceived success, trust, stress, conflict, and documentary routines. (shrink)