In this paper, we report an experimental comparison between widely used combination rules, i.e. sum, product, maximum, borda count and best rank rules. We focus on the behavior of the considered combination rules for ensembles of classifiers exhibiting different performance of recognition. To this end, a simulation method using as input a specified performance matrix (i.e. recognition rates for each position of the true class and rejection rates) is proposed. This simulator generates ensembles of classifier outputs with specified profiles of performance within their list of solutions. Our experimental results tend to show that these profiles of performance are one of the factors that affect the behavior of the combination rules.