Pain pressure thresholds (PPT) were measured at 13 cephalic points bilaterally in 30 headache patients (10 with tension-type headache, 10 with migraine and 10 with cervicogenic headache) and 10 control subjects on three different days. During the sessions, the subjects reported their pain intensity on the right and left sides of the head on a visual analogue scale (VAS). The variability between days was estimated as a coefficient of repeatability (CR = 2 standard deviations of intraindividual differences). The mean CR across all 13 locations was larger in headache patients (2.0 kg/cm2) than in controls (1.68 kg/cm2). Variability (CR) was larger in headache patients as compared to control subjects for 11 of the 13 points (p = 0.02). Reliability was better in controls (intraclass correlation coefficients (ICC) ranging from 0.55 to 0.85) than in headache patients (ICC ranging from 0.43 to 0.78). A moderate negative association between PPT and pain intensity was demonstrated. The intraindividual PPT difference (PPT on the most painful occasions-PPT on the least painful occasions) was negative at 12 of 13 cephalic points (p = 0.003, across-location mean difference: -0.20 kg/cm2). The PPT differed significantly from one day to the next. A part of this variation was presumably related to the circumstances around the procedure; thresholds were lower when the subjects came directly to algometry without any preceding medical examination at all 13 points (p = 0.0002). These results have implications for the planning of future algometer studies. The sample size that is required in studies of headache patients is greater than that in studies of healthy subjects, especially when patients suffer from pain during the PPT session. Particular attention should be paid to circumstances (e.g. preceding medical investigations) around the algometry procedure in order to reduce variability.