Traditional answering methods evaluate queries only against positive and definite knowledge expressed by means of facts and deduction rules. They do not make use of negative, disjunctive or existential information. Negative or indefinite knowledge is however often available in knowledge base systems, either as design requirements, or as observed properties. Such knowledge can serve to rule out unproductive subexpressions during query answering. In this article, we propose an approach for constraining any conventional query answering procedure with general, possibly negative or indefinite formulas, so as to discard impossible cases and to avoid redundant evaluations. This approach does not impose additional conditions on the positive and definite knowledge, nor does it assume any particular semantics for negation. It adopts that of the conventional query answering procedure it constrains. This is achieved by relying on meta-interpretation for specifying the constraining process. The soundness, completeness, and termination of the underlying query answering procedure are not compromised. Constrained query answering can be applied for answering queries more efficiently as well as for generating more informative, intensional answers.