STRING is a database of known and predicted protein-protein interactions.
The interactions include direct (physical) and indirect (functional) associations; they are derived from four sources:
 

Genomic Context		High-throughput Experiments		(Conserved) Coexpression		Previous Knowledge

 
STRING quantitatively integrates interaction data from these sources for a large number of organisms, and transfers information between these organisms where applicable. The database currently contains 1,513,782 proteins in 373 species.

STRING uses orthology information from the excellent COG database.
Up-to-date genomes and proteomes are maintained at Ensembl, GenomeReviews and RefSeq.
STRING references: von Mering et.al. 2007  /  2005  /  2003  /  Snel et.al. 2000.
Miscellaneous: Access Statistics, Robot Access Guide, Medusa Network Viewer, Supported Browsers.
Warning: You are looking at an outdated version of STRING (7.0). For the current release, click here.