We describe Streamer, the query-reformulation component of a data integration system. Given a utility measure and a user query, Streamer uses abstraction-based reenement planning and exploits information on plan independence to produce, in decreasing order of utility, a set of plans that access data sources to obtain answers to the query. We then focus on plan coverage as an important utility measure. We show how to use statistic information about the domain and data sources to estimate plan coverage, and how to incorporate the plan-coverage framework into Streamer. In doing so, we provide the rst method for eeectively integrating the use of quantitative information into the query optimizer of a data-integration system. We present preliminary experimental results suggesting that Streamer runs an order of magnitude faster than brute-force plan-ordering methods, which are the only currently available methods to compute exact plan orderings. Finally, we propose methods to make Streamer scalable to large domains .