The current practice of patient-controlled analgesia has grown from empirical observations. Although several variants of patient-controlled analgesia, bolus doses, infusions, or combinations of both have been suggested, a scientific basis for advocating one variant over the others has been lacking. Most systems have been based on the simplest system, bolus demand, although the use of a combined bolus and background infusion method has theoretical merit. Similarly, a scientific basis for setting the variables of patient-controlled analgesia, drug choice, incremental dose, maximum dose and lockout interval, also has been lacking. Settings for these variables may be rationalised post hoc on the basis of the physicochemical properties and global pharmacokinetic properties of the opioids used but knowledge of these properties has not helped in setting the variables a priori. Foremost, the drug choice should be based on therapeutic index. Knowledge of the regional kinetics of drug (influx and efflux) from brain may provide a more logical basis for setting the patient-controlled analgesia variables but such information can only come from animal experiments. More research is required if patient-controlled analgesia is to become anything but an empirical tool in the quest for improved analgesia in patients.