In 1980 Hoar et al first drew attention to the potential value of job-exposure matrices in the search for new occupational carcinogens.1 They used a matrix to reanalyse data from a case-control study of bladder cancer and found a higher relative risk for the group of jobs entailing possible exposure to aromatic amines than for any of the industrial categories that had been examined in the original analysis. Despite this initial promise, subsequent attempts to enhance hypothesis generating epidemiological sur veys by the application of job-exposure matrices have been largely unrewarding.2 4 One difficulty is the dis crimination of spurious chance associations that inev itably arise when many putative carcinogens are examined simultaneously. One approach to this prob lem is to grade the exposures associated with jobs and look at dose-response relations. When broken down by exposure grade, however, risk estimates are subject to greater sampling variation, and even in relatively large studies known carcinogens, when examined in this way, have failed to show a clear dose response effect. Better results might be obtained with very large data sets in which risk estimates are statistically more stable. To test this idea we have applied a job-exposure matrix to data on 31925 deaths from lung cancer derived from the Office of Population Censuses and Surveys (OPCS) 1971 Decennial Supplement on Occupational Mortality in England and Wales.5 We have included in the matrix several known and sus pected lung carcinogens together with other exposures that are unlikely to cause the disease, and we have concentrated in particular on exposures that are encountered in a range of different jobs since it is for such agents that the matrix method is most likely to offer an advantage over conventional analytical techniques.