The purpose of this study was to classify selective oestrogen receptor modulators based on gene expression profiles produced in breast cancer cells expressing either wtERa or mutant 351 ERa. In total, 54 microarray experiments were carried out by using a commercially available Atlas cDNA Expression Arrays (Clontech), containing 588 cancer-related genes. Nine sets of data were generated for each cell line following 24 h of treatment: expression data were obtained for cells treated with vehicle EtOH (Control); with 10 79 or 10 78 M oestradiol; with 10 76 M 4-hydroxytamoxifen; with 10 76 M raloxifene; with 10 76 M idoxifene, with 10 76 M EM 652, with 10 76 M GW 7604; with 5610 75 M resveratrol and with 10 76 M ICI 182,780. We developed a new algorithm 'Expression Signatures' to classify compounds on the basis of differential gene expression profiles. We created dendrograms for each cell line, in which branches represent relationships between compounds. Additionally, clustering analysis was performed using different subsets of genes to assess the robustness of the analysis. In general, only small differences between gene expression profiles treated with compounds were observed with correlation coefficients ranged from 0.83 to 0.98. This observation may be explained by the use of the same cell context for treatments with compounds that essentially belong to the same class of drugs with oestrogen receptors related mechanisms. The most surprising observation was that ICI 182,780 clustered together with oestrodiol and raloxifene for cells expressing wtERa and clustered together with EM 652 for cells expressing mutant 351 ERa. These data provide a rationale for a more precise and elaborate study in which custom made oligonucleotide arrays can be used with comprehensive sets of genes known to have consensus and putative oestrogen response elements in their promoter regions.