Abstract Pollination requires a flower to remain open for long enough to allow for the arrival of pollinators. However, maintaining flowers costs energy and resources. Therefore, flower longevity, the length of time a flower remains viable, is critical for the outcome of plant reproduction. Although previous studies showed that the evolution of flower longevity depends on the rates of pollen deposition and removal, whether plants should increase or decrease flower life span when the pollination environment is unpredictable has not been explored. Moreover, the common hypothesis that an unpredictable pollination environment should select for increased flower longevity may be too simplistic since there is no distinction drawn between the effects of spatial and temporal variation. Adopting evolutionary game theory, we investigate the evolution of flower longevity under three types of variation: spatial heterogeneity, daily fluctuations within a flowering season and yearly fluctuations between flowering seasons. We find that spatial heterogeneity often selects for a shorter flower lifespan, while temporal fluctuations of fitness accrual rates at both daily and yearly time scales tends to favour greater longevity, although daily and yearly fluctuations have somewhat different effects. However, the presence of correlation between female and male fitness accrual rates seems to have no effect on flower longevity. Our work suggests that explicit measurements of spatial and temporal variation in both female and male functions may provide a better understanding of the evolution of flower longevity and reproduction.