We present here DIA-DB, a web server for the prediction of diabetes drugs that uses two different approaches; a) comparison by similarity with a curated database of anti-diabetic drugs and experimental compounds, and b) inverse virtual screening of the input molecules chosen by the users against a set of protein targets identified as key elements in diabetes. The server is open to all users. registration is not necessary, and a detailed report with the prediction results are sent to the user by email. This is the first public domain database for diabetes drugs.