To determine whether patients should participate directly in detecting adverse reactions to drugs their ability to provide written reports of symptoms experienced during treatment with amoxycillin or trimethoprim-sulphamethoxazole was investigated. When compared with telephone interviews forms on which patients reported events were reliable (the observed agreement with the same statements posed during telephone calls was 85%, κ = 0.56) and valid (sensitivity = 54%, specificity = 94%). Patients were also supplied with forms that invited them to report adverse reactions, and their perceptions were compared with those of a panel of experts, who were informed of all clinical events that had been reported during the detailed telephone interviews. Patients were more conservative than the experts in attributing clinical events to drug treatment. The extent of agreement varied and was notably poor for skin and bowel complaints (κ = 0.13 in each case). The performance of event report forms and reaction report forms as instruments of detection was compared in a hypothetical situation in which the experts' views represented the 'truth' about adverse reactions to a new drug. Event reporting had a higher sensitivity than reaction reporting (42% v 24%) but a lower specificity (58% v 98%). National centres monitoring adverse drug reactions should probably resist pressure to accept reports of reactions directly from the public, but a system based on large scale reporting of events might be valuable in aiding the early detection of symptomatic reactions to new drugs.