Invasion by undesirable plants, such as Cirsium arvense, can constrain attempts to conserve and restore biodiversity in extensively managed temperate grasslands, but control with herbicides can cause environmental harm. We contrasted herbicides with more environmentally sustainable weed control strategies. Six-year, large-scale randomized block experiments were established to determine optimum combinations of grazing management and mechanical or herbicide treatments to control thistles within lowland and upland grazing systems. Factorial combinations of tight vs. lenient grazing in spring and autumn with additional treatments of winter grazing were compared. Thistle control methods were applied in sub-treatments for the first 2 years: cutting twice yearly, herbicide wiping, and cutting followed by herbicide application. Thistle abundance decreased under lenient grazing in spring, autumn and winter at the lowland site, under lenient spring and winter grazing in the uplands and under cattle compared with sheep grazing. Herbicide wiping was the most effective control measure and cutting the least, but effects of all weed control sub-treatments were lost rapidly, so lenient grazing was sufficient to give long-term thistle control. Lenient grazing and herbicide wiping also caused small declines in non-target forb diversity. Control of creeping thistle can therefore be achieved without herbicides because lenient grazing in spring and autumn can decrease thistle populations to sufficiently low levels. Severe infestations can be more rapidly controlled using herbicides, but are better avoided at botanically diverse sites. Ecologically-based weed control strategies have great potential, but require well designed field experiments, which run for sufficiently long periods to allow community-level impacts to develop.