A sustainable approach in the construction industry requires civil engineering professionals with technical and soft skills. Those skills complement each other and facilitate the professional to work effectively in multidisciplinary groups during the development of construction projects. Universities apply collaborative learning methods such as group work (GW) in the classroom to achieve these skills. There is disagreement on the best way to select the members of the GW to achieve their best performance, but it is clear that it should favor the interaction of diverse actors to promote the development of soft skills. A random or criteria-based selection could bring groups of people very close together, leading to the academic homogeneity of GW members and impairing performance and learning. Even the most alert instructors lack information about the closeness of their students, so they rely on their intuition without having tools that allow them to confirm their assumptions or relate them to GW performance. The objective of this paper was to discover the social structures within the classrooms and to identify the groups of people close by trust, knowledge, and informal conversation to relate them to their GW performance. For this purpose, a social network analysis (SNA) was applied to Civil Engineering degree students. In addition, a correlation analysis between SNA metrics and GW grades was performed. The results show that beyond the way in which members are selected, there is a social structure that affects such selection and GW performance. This study presents information that can be used for instructors for a better GW selection that propitiates the development of soft skills in Civil Engineering students.