A method is presented for computing minimal answers of the form ÚA\bigvee {\cal A} in disjunctive deductive databases under the disjunctive stable model semantics. Such answers are constructed by repeatedly extending partial answers. Our method is complete (in that every minimal answer can be computed) and does not admit redundancy (in the sense that every partial answer generated can be extended to a minimal answer), thus no non-minimal answer is generated. The method does not (necessarily) require the computation of models of the database in their entirety. A partitioning of the database into extensional and intensional components is employed in order to overcome the problems caused by the possible non-existence of disjunctive stable models, and a form of compilation is presented as a means of simplifying and improving the efficiency of the run-time computation, which then reduces to relatively trivial processing within the extensional database. In addition, the output from this compilation process has the significant advantage of being immune to updates to the extensional database. Other forms of database pre-processing are also considered, and three transformations are presented mapping a database onto an equivalent positive database, non-disjunctive database, and set of conditional facts.