Estimates of genetic variation and selection allow for quantitative predictions of evolutionary change, at least in controlled laboratory experiments. Natural populations are, however, different in many ways, and natural selection on heritable traits does not always result in phenotypic change. To test whether we were able to predict the evolutionary dynamics of a complex trait measured in a natural, heterogeneous environment, we performed, over an 8-year period, a two-way selection experiment on clutch size in a subdivided island population of great tits (Parus major). Despite strong artificial selection, there was no clear evidence for evolutionary change at the phenotypic level. Environmentally induced differences in clutch size among years are, however, large and can mask evolutionary changes. Indeed, genetic changes in clutch size, inferred from a statistical model, did not deviate systematically from those predicted. Although this shows that estimates of genetic variation and selection can indeed provide quantitative predictions of evolutionary change, also in the wild, it also emphasizes that demonstrating evolution in wild populations is difficult, and that the interpretation of phenotypic trends requires great care.