Why are observed contracts so often incomplete in the sense that they
leave contracting parties' obligations vague or unspecified? Tradition
al answers to this question invoke transaction costs or bounded ration
ality. In contrast, we argue that such incompleteness is often an esse
ntial feature of a well-designed contract. Specifically once some aspe
cts of performance are unverifiable, it is often optimal to leave othe
r verifiable aspects of performance unspecified. We explore the condit
ions under which this occurs, and investigate the structure of optimal
contracts when these conditions are satisfied.