We report on protein-protein interactions in the human proteome identified using a high-throughput two-hybrid screen. The raw network contains over 30,000 unique interacting pairs. We describe bioinformatic analysis to identify a high-confidence core network, evolutionary comparisons to model organism networks, and estimates of the underlying complexity of human protein-protein interactions. The experimental work was conducted by CuraGen Corporation, and the bioinformatic analysis is from an ongoing collaboration between CuraGen and Johns Hopkins.