The excess number of weekly laboratory requests for confirmation of dengue diagnosis over the expected number of requests forecasted by the modified Serfling method is proposed for the surveillance of dengue in French Polynesia, in addition to conventional methods. Retrospective analysis of the seasonal curves of dengue activity related to the number of laboratory requests is described for the years 1982–1987 where dengue type 4 was the only active flavivirus at the time when the forecast was initiated. By using past epidemic data, the probability of,failing to recognize an increase in excess of requests as possibly epidemic was of 13.2% and 5.8%, respectively, when the criterion for epidemic increase was set respectively at 2 and 3 successive weeks during which the epidemic threshold is exceeded. A weekly surveillance was set up prospectively for 1988 using these criteria.