Lithium is widely used in the prophylaxis of episodic cluster headache without formal evidence of efficacy. Placebo-controlled clinical trials are not easy in conditions characterized by frequent severe pain. In this study, it was assumed that lithium would work quickly if at all, and placebo response would be zero. Strict diagnostic criteria excluded uncertain or atypical cases. Patients were male in so-far untreated episodes expected to last for at least 3 weeks more. In a double-blind, placebo-controlled comparison of matched parallel groups, treatment was either slow-release lithium carbonate, 800 mg/day, or placebo. After 7 days, compliance was estimated by tablet count, blood was taken for lithium assay, efficacy was assessed (attacks stopped or substantially improved) and adverse reactions were recorded. The study was stopped after planned sequential analysis of the 27th patient (13 on lithium, 14 on placebo). Estimated compliance was usually but not always good. Plasma lithium levels were mostly in the range 0.5-0.6 mmol/l on lithium, zero on placebo. Cessation of attacks within 1 week occurred in two patients in each group, substantial improvement in 6/14 (43%) on placebo, 8/13 (62%; NS) on lithium. Only minor adverse events were reported. Lithium treatment was therefore associated with a useful subjective improvement rate but the assumptions made at outset had proved wrong. The trial was stopped because superiority over placebo could not be demonstrated. There were lessons for future trials.