By setting out arrays of potted plants of Penstemon strictus, I tested whether freely foraging bumblebee (Bombus spp. ) workers would establish regular foraging routes that reflected the geometry of the array. They did, passing through an asymmetrical array in a pattern that minimized interplant flight distances. After the array was changed to a symmetrical pattern, however, the experienced bees continued to show their previous asymmetrical flight patterns. New bees without experience on the asymmetrical array showed no asymmetry on the symmetrical array. I term this persistence of flight-path geometry “trapline holdover, ” and discuss its implications for the study of animals' learning and foraging behavior.