While collaboration is an important and key attribute for medical students in order to prepare them to perform well in health care teams, how to effectively develop and assess such skills is challenging. The current widespread practice of using Likert-scale questionnaire only to measure the quantity of collaboration at course and/or program level appears to be insufficient to provide an evidence-base for what counts desirable collaborative learning experience. Drawing on research into "student approaches to learning" and "social network analysis," this study investigates differences in collaborative learning configurations amongst 217 Australian medical students. Based on students' learning orientations (i.e., 'understanding' and 'reproducing') and their choice of collaborations (i.e., whether to collaborate or not, with whom to collaborate, and mode of collaboration), the analyses found five configurations of collaborations differing in a number of features. The most desirable collaborative experience was a configuration of collaborations formed by students with an 'understanding' orientation. This configuration revealed a strong tendency towards intensive pair work with measurable differences in how easy and effectively they collaborated. The results of the study not only have practical implications for teaching and curriculum design for collaborative learning, but also have significant implications for assessing students' collaborative learning experiences.