The main function of the World Health Organisation's International Collaborative Programme on Drug Monitoring is to provide a reliable early warning of possible health hazards caused by medicines. Described here is an attempt to devise criteria that would produce a well-founded early signal of an adverse reaction on the basis of reports sent in by national collaborating centres and combined in the WHO database. To reduce the frequency of spurious associations (false-positive signals) it is suggested that publication be delayed until a few case-histories meeting the suggested criteria have been sent in. The criteria were tested retrospectively against early published case-reports on drug-associated agranulocytosis. 19 suspected associations were examined and a signal in the database was defined by there being three or more cases containing stipulated information about the patient and the treatment. The WHO database had reports on all the associations, suggested criteria for a signal being met in 15 instances. This signal was present when the first case was published in 7 instances and within three months of first publication in 1. Moreover, in 3 instances where publication came first the cases presented had been collected by a national drug monitoring centre. The WHO databank has the potential to provide doctors and scientists with signals which then should be evaluated in detail.