Several short-term trials of baclofen for tardive dyskinesia have yielded conflicting results. We evaluated 33 patients maintained on a constant dose of neuroleptic during a randomized 6-week double-blind placebo-controlled trial, followed by a 6-week open-label trial. During the double-blind part, 67% of baclofen-treated patients and 47% of placebo-treated patients showed a reduction of 25% or more in their original dyskinesia score on the abnormal involuntary movement scale. In the open-labeled part, we further evaluated seven of nine baclofen responders and nine of nine placebo nonresponders. We found a 77% baclofen response rate for the original placebo nonresponders, a loss of efficacy in 23% of the original baclofen responders, and an attenuation of response in others. Patients with major depressive disorders were poorer responders to baclofen than those with schizophrenia. Side effects were infrequent and reversible, but dose limiting in several instances. Although baclofen may help some cases of mild tardive dyskinesia, its efficacy may attenuate with long-term administration.