This paper presents an empirical study of the relations between scientific output and collaboration performed on two scales: (1) an individual scale, for members of a study model, and (2) a group scale, for three samples varying in the level of productivity. The rank approach was applied in the preparation of the study model resulting in the selection of a set of the most prolific authors. In the course of that process, multiple authorship problem was solved by a dual approach, consisting of normal count and modified straight count procedures. As shown by the analysis of collaborative patterns, either on individual or on group scales, scientific output is highly dependent on the frequency of collaboration among the same authors. Expressed as the collaboration measure, it might serve as an indicator in comparative analyses of scientific productivity in a given field of science.