Collaboration matters in science.  Yet, while there is intriguing evidence that the organization of scientific collaboration is changing, and that teams are becoming more important, the analysis has been mostly based on studying the results of collaborations as evidenced by co-authorship of publications or patents at the scientist level. There are a number of unanswered questions about the structure of the fundamental unit of scientific production: the project team.   Who works on scientific teams?  What is the role of postdoctoral fellows and graduate and undergraduate students? How do scientific networks of collaboration evolve in response to federal funding decisions? And how do different network structures affect scientific productivity, in both creation of knowledge and diffusion of results? And what is the interplay of gender in the complex mix of team dynamics? We contribute to the literature by using new longitudinal data derived from the STAR METRICS[1] program to examine a new level of analysis in depth: the structure of scientific collaborations and the evolution of scientific project teams and networks. These new data have several features that make them particularly attractive for studying collaboration.  They can be used to examine the network structure of project teams over time, since there is longitudinal information on all participants (and their occupations) in project teams at research institutions as well as information on other inputs; such as the expenditures necessary to support their activities on their projects. They can be used to compare collaborations within and across scientific areas; the scientific topics that are being studied have been summarized using natural language processing and topic modeling techniques.  By linking information about team members, including their gender, with existing data on patents and publications we describe and possibly assess the results of collaborations on the knowledge production and transmission process.  Finally, we make use of both quantitative and qualitative methods to better understand the knowledge process by using both information on scientists and their views on their collaborations..