Many insects foraging for nectar or pollen exhibit flower constancy, a learned fidelity to a particular species of plant that previously provided a reward. Constancy may persist even when alternative flowers are available that provide a greater or less variable reward. This strategy entails more travelling time than one of generalization (visiting all suitable flowers as they are encountered). The consensus at present is that this increase in travelling time is offset by decreases in handling time; switching between flower species incurs a cost in time spent learning to 'handle' the new flower species that is avoided by remaining constant. If this is so, then the optimal strategy should depend upon the density of flower species (and thus the travelling time), with switching occurring below a threshold density of the target flower species. This prediction is tested using the butterfly, Thymelicusfiavus, by analysing foraging patterns under natural conditions. This species exhibited constancy: of 465 visits to flowers 85% were to the same species as last visited. As predicted switches between flower species occurred in response to low encounter rates of the flower species on which the individual had previously fed. However, butterflies ignored the vast majority of suitable flowers that they encountered, even when they were of the species to which they were constant. This casts doubt on explanations for flower constancy as an adaptive strategy that minimizes handling time and maximizes resource acquisition per unit time within learning constraints.