The total number of viable, undamaged seeds released by a plant provides an estimate of female reproductive success (RS). I outline a method for analyzing pollinator-mediated selection based on partitioning female RS into four multiplicative components: number of flowers, pollen received per flower, number of seeds set per grain of pollen received, and number of undamaged seeds per seed set. I measured these fitness components for individual plants in several populations of Ipomopsis aggregata and estimated the life-cycle components of selection differentials. Plants whose flowers had highly exserted stigmata and spent a high proportion of time in the pistillate phase received more pollen per flower, as estimated by tracking dye. Experimental hand-pollinations showed that female RS was strongly limited by low levels of pollen, confirming the results of other studies. In spite of this pollen limiation, within natural populations female RS did not increase with mean pollen received per flower. Of the four multiplicative fitness components, only number of flowers and number of seeds set per grain of pollen correlated positively with female RS. As a result, even though hummingbirds exerted significant selection through female function, net selection was not predictable from pollination success alone. Instead, selection was modified by events during flower production, seed formation, and predispersal seed predation. Net selection on proportion pistillate was stabilizing. Selection on flowering date reversed direction over the life cycle, but net selection strongly favored early blooming. Selection on corolla length varied from year to year, which may help explain the high level of phenotypic variation. The relationships between fitness components differed depending on whether I compared flowers within plants, plants within populations, or population means. Failure to take plant and population identity into account can lead to false detection of selection when none is present.
