In occupational epidemiology, failure to take gender into account in the analyses may contribute to conceal certain issues specific to women's health, such as the carcinogenic risks they may incur in the workplace. The aim of this thesis is to analyze how gender-related biases affect different stages of the process of producing epidemiological knowledge on occupational cancers, in particular in the definition of the scope of investigation, the exposure assessment and the statistical modelling.The first significant bias is the limited number of studies assessing the etiology of occupational cancers in women. Through a systematic review of observational studies (n = 243) indexed on PubMed for lung cancer, we show that carcinogenic risks remain largely under-studied in women as compared to men. Secondly, based on the Giscop93 study, we have compared, at the job level (n= 7 702), an assessment of occupational exposure to chlorinated solvents made by an expert panel to an evaluation performed by a general population job-exposure matrix Matgéné. Our results suggest some disagreements (in the exposure indices assigned) between the two methods, notably according to the degree of feminization of the job evaluated. Thirdly, based on the ICARE case-control study (cases, n = 2 926 (22% were women), controls, n = 3 555 (22% were women)), we show that the odds ratio of lung cancer associated with exposure to chlorinated solvents is modified by sex and occupational category.This body of work has led us to proposals for research practices and new methods for the analysis and control of gender bias in occupational epidemiology studies.