Expression signatures obtained from tumour samples are widely used to characterise tumour subtypes and to provide prognostic information. This commentary, based on a recent paper, questions the scientific validity of such signatures as a means of understanding the underlying biology.