In a disease monitoring system embracing several diseases and communities, alarms indicating a significant increase in disease incidence will frequently occur even though the false alarm rate has been set quite low. An approach is presented by which an alarm is determined as either confirmed or rejected according to data observed subsequent to the alarm. The suggested procedure is expected to weed out a substantial proportion (say about 75 per cent) of the false alarms at the expense of some delay in detecting a true alarm.