In this talk, I will first review the use of exponential family models in social network analysis, by presenting the Exponential Random Graph Model (ERGM). ERGMs are a principled statistical tool to model social networks and draw inference on the complex mechanisms driving the formation of ties in these networks. I will then introduce a similar statistical framework for cross-sectional partitions, that is social groups with exclusive membership. This framework (that we call the Exponential Random Partition Model) draws from stochastic models for networks and partitions (in particular from biology). I will describe its main mathematical properties and show how it can be specified and estimated using case studies. I will also briefly desribe potential links to game theory and coalition formation models.