OBJECTIVE In patients with Alzheimer's disease (AD) with psychosis or agitation that respond to haloperidol treatment, to evaluate the risk of relapse following discontinuation. METHODS In outpatients with AD with symptoms of psychosis or agitation, responders to 20 weeks of haloperidol (0.5-5 mg daily) were randomized to a 24-week, double-blind pilot trial of discontinuation on placebo versus continuation haloperidol. Phase A response criteria were minimum 50% reduction in three target symptoms, and improvement on the Clinical Global Impression-Change (CGI-C) score for psychosis/agitation. Phase B relapse criteria required 50% worsening in target symptoms and on the CGI-C. α = 0.1 was the significance criterion in this pilot study. RESULTS Of 44 patients, 22 patients responded in Phase A. The sum score of target symptoms, and Brief Psychiatric Rating Scale (BPRS) psychosis and hostile suspiciousness factor scores, decreased in Phase A (p's < 0.001). Extrapyramidal signs increased in Phase A (p < 0.01). Of 22 responders, 21 patients entered Phase B, and 20 had at least one follow-up visit. Four of 10 patients (40%) on continuation haloperidol relapsed compared to eight of 10 patients on placebo (80%, χ(2)  = 3.3, p = 0.07). In survival analyses, time to relapse was shorter on placebo than haloperidol (χ(2)  = 4.1, p = 0.04). CONCLUSIONS Haloperidol open treatment was efficacious, and relapse was greater on placebo than with haloperidol continuation. In patients with AD who have psychosis or agitation and respond to antipsychotic medication, the increased risk of relapse after discontinuation needs to be weighed against the side effects associated with continuing the medication.