Rigorous experimental methods suppress differences among study participants (noise) to detect true intervention effects (signals). But suppressing participants' heterogeneity obscures an essential dimension of biological and clinical knowledge. Medicine is therefore ambivalent about the influence of heterogeneity on outcomes and struggles to find ways to take it properly into account in both clinical practice and research. This analysis explores the roots of that ambivalence. Drawing on the evaluation of 2 health care improvement initiatives, this article examines the unique features of improvement that help to understand heterogeneity's influence on study methods, and suggests a variety of ways to assess the effect of heterogeneity on study outcome measures.