A prediction from the herbivore optimization hypothesis is that for every combination of site/habitat type and plant community type there is a grazing intensity that causes a maximum increase in above-ground net primary productivity compared with the ungrazed control. NPP is defined as the rate of change in green, herbaceous biomass per unit time per unit area. We tested this hypothesis in the primary summer range of a growing population of wood bison (Bison bison athabascae) within the Mackenzie Bison Sanctuary, Northwest Territories, Canada. Plots (0.5 x 0.5 m) in graminoid meadows dominated by awned sedge (Carex atherodes) were either clipped at 3 cm, exposed to wood bison grazing, temporarily protected for 3 weeks, or permanently protected. This resulted in the removal of 100%, 0-79%, 0-79% or 0%, respectively, of shoot tissue available to wood bison. NPP of meadows clipped twice at 3 cm in 1986 was the same as control NPP at 5 study sites. In 1987, only the 2 most productive study sites of 1986 were intensively examined: plots clipped once in early summer increased in NPP by 120% and 133% compared to controls; NPP of meadows grazed by wood bison increased by 200% compared to controls at the most productive site, but remained the same as controls at the less productive site. Therefore, the herbivore optimization hypothesis was accepted at the 2 most productive sites in 1987, but rejected at all 5 study sites in 1986. In 1987, the standing crop of dead material was 258% and 142% higher in controls than in grazed plots at the 2 most productive sites. We think this dead material was responsible for the lower NPP observed in control plots.