We investigated the conditions under which plastic responses to density are adaptive in natural populations of Impatiens capensis and determined whether plasticity has evolved differently in different selective environments. Previous studies showed that a population that evolved in a sunny site exhibited greater plasticity in response to density than did a population that evolved in a woodland site. Using replicate inbred lines in a reciprocal transplant that included a density manipulation, we asked whether such population differentiation was consistent with the hypothesis of adaptive divergence. We hypothesized that plasticity would be more strongly favored in the sunny site than in the woodland site; consequently, we predicted that selection would be more strongly density dependent in the sunny site, favoring the phenotype that was expressed at each density. Selection on internode length and flowering date was consistent with the hypothesis of adaptive divergence in plasticity. Few costs or benefits of plasticity were detected independently from the expressed phenotype, so plasticity was selected primarily through selection on the phenotype. Correlations between phenotypes and their plasticity varied with the environment and would cause indirect selection on plasticity to be environment dependent. We showed that an appropriate plastic response even to a rare environment can greatly increase genotypic fitness when that environment is favorable. Selection on the measured characters contributed to local adaptation and fully accounted for fitness differences between populations in all treatments except the woodland site at natural density.