A total of 719 lung cancers were identified in a follow-up of 135,000 Swedish construction workers interviewed during 1971-1974 by occupational health nurses as part of a routine health check-up. These were analysed in a matched case-control study nested within this cohort, to compare different methods of characterizing exposure to asbestos. Self-reported exposure was contrasted with a job exposure matrix (JEM) of five levels of exposure intensity, applied to the job at the time of health check-up. Smoking adjusted odds ratios were computed and the JEM performed better than self-reported exposure, in being able to discriminate high risk subgroups. The same pattern was evident in a parallel analysis of 41 mesotheliomas. However both measures appeared subject to misclassification, and the question put seemed to pick up the use of asbestos cement products more effectively than asbestos insulation products. It is concluded that a simple JEM can be more reliable than a simple question, but that both should be much more detailed, to take account of different types of asbestos exposure and their variation over time.