Standardized mortality ratios (SMRs) for occupational diseases are confounded by health differences between industrial and general populations. In 109 industrial cohorts largely free of work-related mortality, these selection effects were sizable for both malignant and nonmalignant outcomes. All-cancer SMRs were considerably less than 1.0 for many cohorts, and lung cancer was subject to almost as much selection-derived confounding as nonmalignant disease. Standardized proportional mortality ratios (PMRs) (approximated by relative SMRs (RSMRs] were less confounded than SMRs in estimating occupational risk. PMRs appeared to overestimate cancer mortality on average by 6%, while SMRs underestimated by 13%. PMRs underestimated nonmalignant respiratory disease by 16 percent but SMRs underestimated by 39 percent. The sources of confounding, in addition to selection on health status at hire, most likely include social class. SMRs, in the absence of internal population comparisons, would fail to detect both malignant and nonmalignant work-related mortality in many industrial cohorts.