45 patients suffering from severe chronic intermittent headaches were divided into 3 groups matched for sex, and assigned to a double-blind 5-week cross-over design with 3 X 1 g/d metamizole--a mild analgesic of the pyrazolone type--a placebo, or a no-treatment control condition. For each of the 6 sessions (t0-t5) signal detection theory parameters d' and log beta for assessment of electrical pain perception and headache ratings on the preceding treatment period were obtained. At t0 all patients were examined via a personality inventory. There were no significant drug effects on headache and signal detection theory parameters, but a clear decrease of the discrimination index d' and of clinical pain over time, regardless of the mode of treatment. No relationship between pathological and experimentally induced pain could be demonstrated. There was no significant negative correlation between response bias of judgement of stimulus intensity (log beta) and neuroticism.