Twenty-two patients with chronic pain of malignant or nonmalignant origin were given intravenous morphine by patient-controlled analgesia. A prestudy judgment was made from the characteristics of the pain as to whether it was nociceptive or neuropathic. Analgesic efficacy was assessed by a nurse-observer; adverse events were noted and plasma morphine and metabolitie concentrations measured. Three categories of opioid response were distinguished. Good responders obtained > 70 mm relief on the visual analogue scale, with minimal or manageable adverse events. Moderate responders obtained < 70 but > 30 mm relief with more problematic adverse events, and poor responders had < 30 mm relief with troublesome adverse events. This method for the study of opioid sensitivity allowed a wide dosage range to be studied. The simultaneous analgesic and adverse event measurements showed that the spectrum of observed response was wide, and response category could be judged for the majority by 4 h. In those with poor or moderate response, adverse event severity limited further dose increment. The relationship between pain characteristics and response showed that some pains judged to be neuropathic had a good response to opioid (5/13), and some pains judged to be nociceptive did not (5/14). The study suggests that the pattern of response is not as black and white as the prediction of good response from nociceptive pain and poor from neuropathic pain would suggest, although nociceptive pain was more likely than neuropathic pain to show a good response. For the moderate responders opioid titration may, in the absence of other effective treatments, be useful, but the analgesic endpoint may not be totally satisfactory. The method provides an operational definition of opioid sensitivity.