Dose-response curves were constructed for intrathecal morphine (M), oxymorphone (OM), hydromorphone (HM), diamorphine (DM), 14-hydroxydihydromorphine (OHM), oxycodone (OC), hydrocodone (HC) and fentanyl (F). Intrathecal catheters were placed in 69 rats under halothane/N2O anaesthesia. After recovery, baseline hot plate and tail flick latencies were measured, and a dose of opioid was given. Hot plate and tail flick latencies were assessed at 5, 15, 30, 60, 90, 120 min and then hourly until they returned to within 25% of baseline. Response latencies were converted to per cent of maximum possible effect (% MPE) and the area under the % MPE X time curve was taken as the response. This measure includes information about both potency and duration of action. Each rat received 3 opioids and saline at intervals of 2-3 days. On a fifth occasion, the animal's first treatment was repeated. Each opioid was studied over an 8-fold dose range. Results of both hot plate and tail flick were best described by a model including log(dose), a component due to development of tolerance over the 5 experimental days, and an among-rat variation term. In the hot plate test, doses equieffective in producing a response (AUC) over the dose range studied were in the order OHM less than OM less than HM less than M less than F less than DM less than HC less than OC. Slopes of the log(dose)-response curves were similar for all drugs except OHM, which had a steeper slope. A model is proposed in which hot plate and tail flick latencies are prolonged while CSF concentrations of a drug are above its minimum effective concentration, and drug is cleared from the CSF by a first-order process, possibly uptake into the spinal cord and removal via the blood. This model predicts that log(dose)-response curves will be linear, as was observed, with slopes inversely proportional to the rate constant for clearance from CSF. According to this model the steeper slope of the OHM log(dose)-response may be interpreted as indicating slower clearance from CSF. OHM has the lowest octanol/pH 7.4 buffer distribution coefficient (0.34) of all opioids studied, possibly leading to a lower rate of uptake into the spinal cord.