We tested the impact of three features of a job-exposure matrix on risk estimates in a case-control study that evaluated the association of methylene chloride and astrocytic brain cancer. These features were probability of use of the agent; the consideration of decade of predominant use of methylene chloride within each occupation; and the use of a more specific industrial-occupational coding system. We compared the risk estimates obtained with and without these features. The introduction of each feature had a striking effect on the estimate of relative risk. The odds ratio ranged from 1.47 with none of these features, to 2.47 with high probability of exposure within industry and occupation, to 4.15 with high probability of exposure and specific industrial-occupational coding, to 6.08 with the three features together. These results indicate that the degree of exposure misclassification can be reduced by the introduction of these features into the job-exposure matrix.