Many countries now have epidemiological surveillance systems using health services-based indicators that allow detection of influenza epidemics. However, there is no accepted criterion for defining an influenza epidemic. An epidemic criterion has been developed, based on a time-series analysis of health services-based indicators collected on a weekly basis by a surveillance network implemented in the Paris region since 1984: the Groupe Régional d'Observation de la Grippe (GROG). For each new season, an epidemic threshold is independently defined for each health services-based indicator as the upper limit of the one-sided confidence interval of the expected value calculated from the weekly differences between the observed number of events and those predicted by a SARIMA model fitted on the non-epidemic data of previous seasons. Epidemic criteria for influenza A and B are then defined from the combination of both viral indicators and epidemic thresholds of individual health services-based indicators. Among health indicators, sick-leave data collected from GP's or the Health Insurance system, emergency home medical visits, and influenza-like-illness reported by GP's are the most sensitive indicators for the early recognition of epidemics. The exceeding of the above mentioned thresholds combined with virological data allows the specific detection of influenza A or B epidemics. This time-series method of analysing surveillance data provides early and reliable recognition of these epidemics.