Plant–pollinator interactions are potential drivers of evolution in floral traits. Because nectar chemical composition is known to mediate both plant–pollinator interactions and plant reproductive success, it can be expected that chemical composition of nectar is subjected to strong pollinator-mediated selective forces. However, the extent of natural selection on different nectar components has not been studied so far. Using the Lepidoptera pollinated fragrant orchid Gymnadenia conopsea as a model species, we used high-performance anion-exchange chromatography (HPAEC) to characterize the sugar and amino acid composition of floral nectar in three calcareous grassland populations of G. conopsea. We then measured phenotypic selection on nectar composition and on other plant and floral traits through applying both linear regression and structural equation modelling. We demonstrate phenotypic selection on plant height, inflorescence height and on specific nectar amino acids, whereas spur length, total sugar and amino acid concentration were not direct targets of selection. Chemical nectar composition is thus indeed under selective pressure but nectar amino acids are much more important to fitness of G. conopsea, as compared to nectar sugars. Furthermore, as we found no evidence of selection on the total amino acid concentration, it is unlikely that amino acids increase pollinator attraction because they are a pollinator nitrogen source. To further unravel the evolutionary ecology of floral nectar, behavioural experiments with pollinators exposed to different nectar components and studies experimentally identifying the selective agents are recommended.