Pollinators may generate selective pressures that affect covariation patterns of multiple traits as well as the mean values of single floral morphological traits. Berg predicted that flowers pollinated by animals whose morphology closely matches the flower's shape will be phenotypically more integrated (tighter correlation of flower traits) than will flowers pollinated by animals not closely fitting the floral morphology. We tested this hypothesis by comparing, in the Strait of Gibraltar region (south Spain, northern Morocco), populations of Narcissus papyraceus that have geographical differences in pollinator faunas. Long-tongued, nectar-feeding moths dominate the pollinator faunas of those populations close to the Strait of Gibraltar, whereas short-tongued, pollen-feeding syrphid flies dominate in peripheral populations farther from the Strait. Populations pollinated by moths and flies differed in the mean values of several floral traits, consistent with the evolution of regional pollination ecorypes. Populations pollinated by moths showed stronger intercorrelation (floral integration) than populations pollinated by hoverflies. Moth-pollinated populations also showed less variation in flower traits than vegetative traits, and this difference was stronger than in fly-pollinated populations. Thus, the pattern of differences in the phenotypic architecture of the Narcissus flowers is consistent with the hypothesis that populations have responded to different selective pressures generated by different pollinators. These data also supported most of the specific predictions of Berg's hypotheses about integration and modularity.