A great deal of recent research has focused on the problem of selecting differentially expressed genes from microarray data (?gene selection?). Recent theoretical work has shown that the effectiveness of a gene selection algorithm can be captured as a probability called ?selection accuracy?. Unfortunately, in practice, there tends to be relatively little known about the very features upon which selection accuracy depends, making it difficult to choose a suitable method. In this paper we present a ?consistency analysis? which allows the inference of posterior distributions over selection accuracy from data. The utility of our approach lies in the fact that it can be used to assess gene selection algorithms in a practical but principled manner, and thus choose an appropriate method for given experimental data.