Spatial patterns are important cues for flower detection and recognition by nectar–feeding insects. Pattern vision has been studied in much detail in bees and flies but rarely in butterflies and moths. In this paper, I present a first proof of pattern–learning abilities in a moth, and discuss reasons for the limitations to their pattern learning. The diurnal hawkmoth Macroglossum stellatarum spontaneously prefers patterned to uniformly coloured stimuli but can be trained to choose the uniform stimulus. By contrast, experience does not override the innate preferences for radial over tangential patterns, and for tangential over striped patterns. These results do not reflect bad visual discrimination but rather a lack of learning ability and motivation to abolish innate preferences. I propose that radial and tangential flower patterns are good predictors of nectar reward, a condition under which learning is unlikely to evolve. These patterns serve not only as cues for flower detection but also as guides to the reward. Hovering pollinators strongly depend on these guides and should therefore: (i) have rigid pattern preferences; and (ii) not be motivated to abolish these preferences as easily as their innate preferences for colours.