Microarrays have quickly been established as an efficient tool for gene expression profiling. In this paper we consider the detection of differentially expressed genes with replicated measurements of expression levels of each gene under each condition. Several authors mentioned that data from microarrays are often not normally distributed, even when suitably preprocessed. Consequently, nonparametric tests, such as the Wilcoxon rank sum test and the Fisher-Pitman permutation test, were recommended. As a further powerful nonparametric test we propose the Baumgartner-Weiž-Schindler (BWS) test. However, when the population variances are unequal a significant result in these tests does not necessarily provide evidence for a difference in location. Note that, in data from microarray experiments, heterogeneous variances are common. Therefore, we suggest a two-stage procedure. If the BWS test applied in stage 1 is significant, a test for a difference in location only is carried out in stage 2. A bootstrap test based on the Welch t statistic can be used in stage 2. However, we demonstrate that a rank-based test recently proposed by Brunner and Munzel (2000) is more powerful.