On both artificial flowers in the laboratory and certain plant species in the field, bumblebees often closely approached flowers and then departed without probing for nectar. In laboratory experiments where nectar rewards were associated with subtle visual or olfactory cues, bumblebees approached and avoided non-rewarding flowers. Flowers that bees entered and probed for nectar contained rewards much more frequently than predicted by chance alone. When there were no external cues associated with nectar content, bees visited rewarding flowers by chance alone, provided rewarding flowers were not spatially clumped. In the field, bumblebees approached and rejected a large proportion of dogbane flowers and red clover inflorescences. On both species, flowers or inflorescences probed by bees contained more nectar than those rejected by bees or those that I chose at random. On fireweed and monkshood, bees rarely or never approached and rejected healthy-looking flowers. Predictions generated by an optimal foraging model were tested on data from four bumblebee species foraging on red clover. The model was highly successful in qualitatively predicting the relationship between handling time and proportion of inflorescences rejected by individual bees, and the relationship between threshold nectar content for acceptance by bees and average resource availability. Thus, bees appeared to use remotely perceived cues to maximize their rates of nectar intake.