Strong sexual selection promotes population viability and evolvability, but sexual conflict can offset such benefits. Male harm adaptations leading to pre-copulatory (i.e. harassment) and/or copulatory (i.e. traumatic insemination) harm to females are taxonomically widespread, depress population growth, and can affect the dynamics of adaptation and evolutionary rescue, but we largely ignore what factors modulate their evolution. Here, we show that temperature drives the divergent evolution of male harm in Drosophila melanogaster, a model species in the study of sexual conflict that exhibits both intense harassment of females and copulatory harm via seminal fluid proteins (SFPs). After 30 generations of experimental evolution of wild flies under cold (20±4ºC), moderate (24±4ºC) and hot (28±4ºC) thermal regimes mimicking natural conditions in the wild (average ± circadian range), we characterized overall levels and thermal plasticity of harm and male harassment behaviours, as well as the seminal proteome. Quick local adaptation was evidenced by higher levels of male harm at evolved (vs. non-evolved) temperatures. Furthermore, pre-copulatory vs. copulatory mechanisms of harm responded divergently to temperature. Male harassment was substantially lower and less plastic in lines evolved at cold temperatures, while the seminal proteome of lines evolved at warm temperatures was characterized by the differential expression of SFPs. Such quick divergent evolutionary responses suggest high levels of standing genetic variation in the population of origin. Accordingly, we derived isogenic lines from the ancestral wild population and show strong quadratic GxE interactions for male reproductive success across studied temperatures. Our results suggest: a) that temperature can be key to understand past and future (e.g. global warming) evolution of sexual conflict and the net effects of sexual selection on populations, and b) that natural temperature fluctuations can maintain high levels of standing genetic variation for male harm traits, contributing to resolve the lek paradox.