We compared the role of grasses and forbs in managed grassland under nutrient-poor conditions via the experimental removal of one of these two species groups and an evaluation of the removal effects over a 5-year chronosequence (beginning 4 years after removal start), focusing on above-ground biomass and community composition. There was evidence for the complementarity of grasses and forbs, as shown by better relative performance of both groups in the mixture, measured as the proportional deviation of their above-ground biomass from that predicted using single-group plots. Relative performance of forbs was better than that of grasses and did not change significantly throughout the experiment. Relative performance of grasses decreased, reaching a minimum value of 0·03, but then increased to 0·71, in parallel with increasing legume proportion in the above-ground biomass (from 0·9 to 5·1%). The botanical composition of the grass group did not respond to forb removal. Among the forbs, subordinate species sensitive to light competition became more abundant after removal of grasses; the forb species richness increased on average by 1·6 species per plot. Our results show that the composition and diversity of the forb group is affected by the presence of grasses, and the presence of forbs in a grassland community assures higher temporal stability in above-ground production. Under increasing depletion of soil nutrients, the grasses derive substantial benefits from the presence of legume species and possibly also from the other forb species. The study supports the need for realistic, long-term experiments to better understand plant community dynamics.