The reason for the wide variation in relapse rates in trials of the withdrawal of anticholiner-gic agents in patients maintained on neuroleptics is unclear and could be due to bias introduced through unblinding. Forced-choice guesses of whether patients had been receiving active or placebo medication were made in a clinical trial. Correctness of guessing was correlated with measures of Parkinsonism. Raters' guesses were better than chance expectation and correlated with ratings of Parkinsonism. Nurses' guesses correlated with their own global measure of side effects but not with raters' measures. Patients' guesses were no better than chance. Unblinding of raters is a significant factor in anticholinergic withdrawal studies and preconceived notions of the value of anticholinergics could therefore be affecting the results of trials, helping to produce the wide variation in relapse rates.