Diagram recognizers need to be aware of the notational conventions used in diagrams. These conventions are complex, and difficult to encode in computational form. Diagram generators are a readily-available source of information about notational conventions. We discuss methods for exploiting diagram generators to construct or improve diagram recognizers. In particular, we discuss our experiences in using Lime (a generator for music notation) to correct recognition results present in the output of Musitek MIDIScan (a recognizer for music notation).