Proteins regulate biological processes through a complex network of dynamical interactions. Protein-protein interfaces, e.g., those involved in signaling complexes, constitute increasingly important therapeutic targets. We present a new method, JET2, for accurately predicting protein interfaces at large scale (Laine and Carbone, 2015). Contrary to machine learning approaches, JET2 combines in a rational and very straightforward way three sequence- and structure-based descriptors of protein residues: evolutionary conservation, physico-chemical properties and local geometry. The implemented strategy yields very precise predictions for a wide range of protein-protein interfaces and discriminates them from small-molecule binding sites. Beyond its predictive power, the approach permits to dissect interaction surfaces and unravel their complexity. JET2 is useful to predict and learn about the different partners of a protein. It allows to identify multiple recognition patches within large protein interfaces and characterize their different properties and origins. We show how the analysis of the predicted patches can foster new strategies for protein-potein interaction modulation and interaction surface redesign. JET2 was applied on more than 20 000 protein structures, representing the non-redundant set of non-homologous chains extracted from the Protein Data Bank. The generated knowledge base will soon be made accessible to the community (Ripoche et al., 2015). JET2 is implemented as a fully automated tool, freely available at www.lcqb.upmc.fr/JET2. E. Laine, A. Carbone. The local geometry and evolutionary conservation of protein surfaces reveal the multiple recognition patches in protein-protein interactions, under revision, 2015. H. Ripoche∗, E. Laine∗, N. Ceres, A. Carbone. JET2 Viewer: a web server predicting multiple protein-protein interaction sites for PDB structures, submitted, 2015.