In 1997, the Legionnaires' disease surveillance system was strengthened following a study that had estimated the completeness of mandatory reporting (MR) in 1995 to be 10%. The objective of this survey was to assess the impact of this measure on the sensitivity and representativeness of the surveillance system in order to propose potential improvements and to determine trends in the incidence of Legionnaires' disease in France over time. Methods: A mail survey was conducted in 1999 among 413 public hospital laboratories to determine Legionnaires’ disease diagnostic practices and the number of cases diagnosed in 1998. These data were cross-referenced with mandatory reporting data and with cases reported by the National Reference Center (CNR). Cases were classified as confirmed (isolation of Legionella, fourfold increase in antibody titer with a second titer of at least 128, positive direct immunofluorescence, presence of soluble urinary antigen) or possible (single elevated titer > 256). A three-source "capture-recapture" method using a log-linear model that accounts for interactions between sources and heterogeneity variables was used to estimate the total number of Legionnaires' disease cases diagnosed in 1998. Results: Among the 288 (70%) responding laboratories, 42% used at least one diagnostic method for Legionnaires' disease. The majority of university hospital laboratories used at least one method (84%). Only 36% of laboratories used the urinary antigen detection test or planned to introduce it in 1999. In the capture-recapture study, a total of 715 cases of Legionnaires' disease were identified across the three sources after removing duplicates (DO=370, CNR=384, LAB=385). The best log-linear model included an interaction between the two sources DO-CNR and DO-LAB. This model estimated the number of Legionnaires' disease cases diagnosed in 1998 at 1,124 [95% CI: 973–1,275], which is twice as many as in 1995. The completeness of the DO was 33% overall [95% CI: 29–38] in 1998. For cases diagnosed by culture, it was 71% [95% CI: 68–73]. The estimated incidence of Legionnaires' disease, after accounting for completeness, rose from 0.9 per 100,000 inhabitants in 1995 to 1.9 cases per 100,000 in 1998. Conclusion: The estimated number of Legionnaires' disease cases doubled between 1995 and 1998. The introduction of the urinary antigen detection test may have led to improved diagnosis of the disease and thus an increase in the number of cases. The completeness of mandatory reporting improved threefold between 1995 and 1998, from 10% to 33%. Several factors may explain this increase in completeness: the impact of the strengthened surveillance system implemented in 1997 and during the 1998 World Cup, and the active collaboration between the CNR and the InVs. The increase in the number of cases and in the completeness of mandatory reporting can also be explained by increased awareness among physicians regarding the disease and case reporting, following the facilitation of biological diagnosis and major outbreaks. However, it cannot be ruled out that the incidence has increased. In the current debate on the risks associated with the presence of Legionella in public or private buildings, our study helps assess the true burden of the disease and illustrates its socioeconomic impact. Despite an improvement in the completeness of mandatory reporting, it remains clearly insufficient for the early detection of epidemic outbreaks and the monitoring of trends. Furthermore, the CNR does not receive all strains isolated in France. Once again, various measures will be proposed (information brochures for healthcare professionals, CNR training for laboratories, etc.) to improve case reporting, increase the number of laboratories using the urinary Ag detection test, and encourage testing for the bacterium and the submission of strains to the CNR.