Realized paternity should decelerate with increased allocation of resources to male function for hermaphroditic plants. If these diminishing returns result from the pollination process, total pollen dispersal would be maximized by restricting removal by individual pollinators and using the services of all available pollinators. The optimal restriction of removal by individual pollinators should depend on pollinator availability, the deceleration in dispersal, and the pattern of pollen removal during succeeding visits to a flower. In this study I measured pollen removal from six species (Aconitum delphinifolium, Aralia hispida, Lupinus sericeus, Mertensia paniculata, Pedicularis bracteosa, and P. contorta) to determine the extent to which removal is restricted and to quantify the removal pattern during repeated visits. For all species except Lupinus, I collected the pollen left in a flower after 1—4 visits by freely foraging bumble bees (Bombus). Pollen removal was estimated by substracting the pollen remaining in a visited flower from total pollen production by an adjacent unvisited anther or flower on the same plant. The pollen—dispensing mechanism of Lupinus allowed me to measure pollen removal directly. Much interspecific variation in pollen removal depends on the likelihood of contact between pollinator and anthers. Bumble bees removed a median of 50—80% of the available pollen during first visits to species with exposed anthers. In contrast, only 19° was removed during the initial manipulation of lupine flowers, which present pollen indirectly. Within species, removal increased with visit duration; it did not additionally depend on the number of visits involved. Removal typically changed in direct proportion to differences in pollen availability. Pollen size also affected removal, but it is unclear that this effect is direct. In most cases, removal was unaffected by the species or caste of bee involved. Succeeding visits to a flower removed either a fixed or, more commonly, a declining proportion of the pollen remaining in a flower. A model of pollen dispersal indicates that a constant removal proportion would maximize dispersal when the number of visits a flower receives varies little. In contrast, diminishing proportional removal would allow plants to maintain pollen dispersal when the frequency of pollinator visits is uncertain.