The aim of this paper is to provide a unitary approach to evaluating the performance of expert system shells. Since any expert system shell has to meet a series of requirements to assure good performance, the authors gather the opinions and expertise of a number of professionals with some experience with expert systems and propose a set of evaluation criteria as a convenient framework for shell evaluation. In addition, based on their estimated relative importance, in a survey of users of expert systems and knowledge engineers, a taxonomy of the considered evaluation criteria is obtained. The paper could be a good guide in identifying capabilities and potential shells and comparing different expert system shells.