The genetic basis of the dry-wet season polyphenism of wing pattern in response to temperature shown by Bicyclus anynana was studied, using a split-family design over four temperatures. Reaction norms crossed, but were only linear in the three highest temperatures, and only when larval development time was used as the environmental axis. Significant full-sib additive variances (VA) and heritabilities (h2) for plasticity were found using slopes of reaction norms in a bootstrap procedure. Heritabilities were lower in intermediate temperatures, mainly due to differences in the residual variances (VR). There was no clear trend in VA across temperatures, contrary to the expectation that VA would have been depleted by natural selection at the extreme temperatures and not depleted at the intermediate temperatures which occur less frequently in the field. Unpredictability in the onset of the following season at intermediate temperatures might lead to selection for diverse flresponses resulting in relatively high VRs. Theoretical models linking reaction norms to genetic parameters in separate environments were difficult to apply in this study, particularly because they are based on the assumption that VRs are constant. However the reaction norm approach combined with quantitative genetics provided a valuable insight into the evolution of the observed polyphenism.