There is a long-standing controversy about the health effects of fumes and vapors generated during paving and waterproofing with bitumen. Such emissions may be carcinogenic. To address this question, a historical multicentric cohort of asphalt workers was assembled by in eight countries. Assessment of historical exposures to known and suspected carcinogens (bitumen, organic vapour, coal tar, respirable silica, diesel exhaust, and asbestos) in the cohort became the main objective of this dissertation. First, published reports on exposures in the industry were reviewed. They provided some insight into the identity of factors that influence exposure to bitumen among road construction workers. However, the data available through published reports have limited value in assessing historical exposure levels in road construction industry. Next, a database of European asphalt individual workers' exposure measurements was created. The database allowed retrieval and consistent coding of 38 data sets, of which 34 have never been described in peer-reviewed scientific literature. It was used to create statistical models of bitumen fume, organic vapour and polycyclic aromatic hydrocarbons exposure intensity among paving workers, the most numerous bitumen-exposed group in the cohort. Individual exposure measurements from pavers (N=1581) were available from all countries enrolled in the study. Statistical models indicated that for paving workers, exposure intensity could be reconstructed quantitatively on the basis of time period and production characteristics. Alternative exposure assessment models were explored in a subset of data with more detailed contextual information in order to see whether they might have produced more optimal exposure grouping for epidemiological analyses. It was observed that on the basis of the available data and given constraints of retrospective cohort study design, we could not produce more uniform exposure groups by employing different exposure models. Internal validity of statistical models was assessed in a cross-validation procedure. External validity was evaluated using data not used to develop the original models. Validation of the models increased our confidence in their applicability to exposure assessment. Next, the statistical models were applied to the creation of an exposure matrix: the link between exposure estimates and actual working conditions in studies companies. Procedures were also developed for semi-quantitative estimation of exposure for (a) cohort members not employed in paving and (b) agents for which few or no exposure measurements were available. Production characteristics in the companies enrolled in the study were ascertained via a questionnaire. Our approach produced a data-driven exposure matrix that can be challenged in future studies and easily re-estimated. Performance of the exposure matrix was evaluated in analysis of the hypothesized relationship between bitumen fume exposure and lung cancer mortality. Constructing quantitative exposure indices appeared to have been justified because (a) the healthy worker effect confounded any associations between duration of exposure and lung cancer mortality and (b) we identified statistically significant associations between average bitumen fume exposure and lung cancer risk (after adjustment for coal tar co-exposure) that require further investigation. Many of the unresolved issues in the cohort analysis may be remedied in a nested case-control study.