Much of the high-quality information available on the web and on various intranets is stored in databases. Yet most people who need access to this information do not know how to query databases using SQL. Querying by filling out forms (of the sort found on many web pages) severely restricts the set of questions a person can ask. Thus, there is great value in Question-Answering Interfaces to Databases (QADBs), which allow people to pose natural language questions. Yet QADBs are only usable if they map natural language questions to SQL queries correctly. In this paper, we introduce a theoretical framework for reliable QADBs, which is the foundation for the fully implemented PRECISE QADB.We report on experiments testing PRECISE on several hundred questions drawn from user studies over three benchmark databases and show that it compares favorably with other QADBs.