Background. - In France, cancer incidence is measured by departmental cancer registries, which cover 15% of the population. Cancer incidence is estimated at the national and regional levels using mortality data by extrapolating the observed incidence-to-mortality ratio from the departments covered by a registry. The use of the incidence-mortality ratio at the departmental level is more uncertain. The objective of this study is to produce departmental estimates of colorectal cancer incidence in France, based on the ratio between the number of new cases and the number of surgical hospital stays recorded in hospital databases. Methods. - This ratio was studied over the 1999–2003 period in the 13 departments covered by a registry. For each sex separately, the number of new cases was analyzed in relation to the number of surgical stays for colorectal cancer resection (colectomies, excisions, pelvectomies) using a Poisson model. Age was included in the model as a fixed effect and the department as a random effect. The model’s ability to predict incidence was tested using cross-validation. The model was then extrapolated to estimate departmental incidence. Results. - In departments covered by a registry, cross-validation demonstrated the model’s good predictive ability, with the exception of one department for men where the difference between predicted and observed incidence reached 10%. Across all departments, the estimated incidence rates, standardized to the global population, ranged from approximately 29 to 44 per 100,000 among men and from 17 to 27 per 100,000 among women. Incidence did not show a clear geographic gradient. Conclusion - Within the registries, the incidence predicted by cross-validation was generally close to the observed incidence. The inclusion of multiple hospital stays per patient certainly represented a minor source of error in these estimates. In fact, our selection included only 2% of multiple hospital stays, with no geographic variations, in 2002 and 2003, the years for which patient chaining was available in the hospital databases. The estimated incidence rates showed moderate geographic variations. The accompanying prediction intervals must be taken into account.