Expression profiling has been extensively applied to the study of breast cancer and undoubtedly is changing the way breast cancer is perceived. Over the past few years, several groups have described prognostic "signatures" (gene lists) that are purported to be more accurate prognostic factors than well established clinical and pathological features. In addition, cDNA and oligonucleotide microarrays have also been used to devise predictive "signatures" in the setting of neoadjuvant chemotherapy setting. However, it seems that the enthusiasm with this new technology has led most of us to turn a blind eye to some serious methodological problems which are evident in landmark papers on breast cancer expression profiling. These issues include small and biased cohorts of patients, inappropriate statistical analysis and lack of thorough validation of the technology. In this review, we critically revisit the most relevant cDNA microarray studies on breast cancer prognosis and prediction published to date. Although the results are promising, further optimisation and standardisation of the technique and properly designed clinical trials are required before microarrays can reliably be used as tools for clinical decision making.