Reed canary grass (Phalaris arundinacea) invades wetlands, forms monotypes, and resists control efforts, suggesting that strong feedbacks sustain its dominance, as in the alternative states model. In nine field experiments, we tested the hypothesis that applying a graminicide (sethoxydim) for three years would progressively reduce Phalaris abundance, and that seeding sedge meadow species (except grasses) would reestablish native plant dominance. The graminicide prevented Phalaris from flowering, reduced its height by 50% and reduced its cover, often to less than 40%. However, only two of the nine sites showed progressive declines over the three-year experiment. The first setback was that Phalaris recovered annually in nearly all treatment plots. A second setback was that seeding did not reestablish sedge meadow. In five sites, unseeded plots had similar numbers of native species as those seeded with either forbs, forbs and graminoids, or graminoids. In four formerly agricultural sites, however, non-native weeds increased in species richness and cover (a third setback). In only one site did the graminicide's effect on Phalaris allow native species to increase in number and cover. But short-term gains were not long-lasting. In year four, three sites that developed high native-species cover were again strongly dominated by Phalaris (a fourth setback). The feedbacks that sustain this invader include resistance to the graminicide aboveground and rapid and robust regrowth from rhizomes and seeds belowground. The weak effect of this graminicide was a surprise; hence, we recommend stronger management actions to control Phalaris.