Benzene, an early “known” carcinogen as defined by the International Agency for Research on Cancer (IARC), opens a window where we can directly see the interaction of laboratory toxicology and epidemiology against a regulatory background. That view can inform consideration of laboratory and human data for chemicals less widely recognized as dangers. I​n 2002, the European Environment Agency published an extensive series of case studies in support of the precautionary principle under the rubric “Late Lessons from Early Warnings.” It includes a history of benzene written by Peter Infante, an epidemiologist who developed key data on benzene and also served for years on the standards team for OSHA. Back in 1977, a time when OSHA was allowed to update standards, the agency issued an emergency temporary standard reducing the benzene PEL from the then recently established 10 ppm to 1 ppm. OSHA was spurred by the first good cohort mortality study of benzene, coauthored by Infante. OSHA’s policy at the time was that exposure to carcinogens should be limited to the lowest feasible level, which OSHA determined to be 1 ppm. After the American Petroleum Institute got the 5th Circuit to stay and then flush the 1 ppm PEL, the venue moved to the Supreme Court. The resulting 1980 “benzene decision” flushed (for a time) the 1 ppm limit and established the concept of “significant risk.” Significant risk was vaguely suggested to be somewhere between 1 in 1,000 and 1 in 10 million. Dozens of papers addressing the risk rate test and its morality were published. The danger of this quantitative approach is risk creep—if one death is okay, what about two, or three, or 100? In 1987, with more data (including the bioassay discussed below) and more analysis, OSHA reissued the 1 ppm limit and made it stand up. OSHA’s analysis predicted 10 in 1,000 deaths at 1 ppm. It was feasibility limited. The 1 ppm limit, set in the Reagan Administration, is pretty much the international mode; a few countries and ACGIH are at 0.5 ppm, and there’s a NIOSH REL at 0.1 ppm. I’d argue the NIOSH REL is the protection the legal and risk analysis process would get us if it were free of overt political interference obstructing OSHA rulemaking. The EPA reference concentration (RfC) for benzene is about 0.01 ppm, and the 1 in 1,000 risk rate is 0.15 ppm (unadjusted for duration), which demonstrates the divergence of occupational limits from modern risk estimates based on occupational data. As an aside, had OSHA decided on 1 mg/m3 as the numerological template for “lowest feasible” rather than 1 ppm, the OSHA limit would be three times more protective. SCIENTIFIC SUPPORT Among the new data for the 1987 rule were the results of an NTP bioassay published in 1986. These data have importance far beyond benzene, where human carcinogenicity has been established, relating site concordance between bioassay and results in people. The United Auto Workers, where I was director of safety and health, did not much participate in the 1978 OSHA rulemaking, but the 1987 proceeding provided a second chance. My entry point was the NTP bioassay, to which few paid attention. Up to this publication, through 1982, laboratory evidence for the carcinogenicity of benzene was “limited,” according to IARC. The bioassay provided “clear” evidence for carcinogenicity in rats and mice of both genders at oral dose levels in the ballpark of the old OSHA standard (10 ppm). At the time, the issue was concordance of dose levels, which I presented. However, in contrast to leukemia, the tumor sites in rats were glandular tumors and skin tumors at various sites. For mice, these same sites presented, as well as lung tumors, but also lymphomas. The important observation is that in people we see leukemia, but in the laboratory we see a variety of sites; related hematopoietic tumors appear only in mice, not rats. Two qualitative risk assessment conclusions follow. First, the tumor site in the laboratory species doesn’t necessarily predict the tumor site in people; carcinogenic potential isn’t necessarily site specific. Second, duplication of the tumor site from people to the laboratory isn’t needed to validate the biological plausibility of the epidemiological observation. A null laboratory result has little weight unless it’s from a lifetime study (not two-year) at the maximum tolerated dose. This bioassay also impacts assessment of formaldehyde, shown through epidemiology to cause leukemia in people. Some have disbelieved the epidemiologic findings, advancing among other reasons the apparent absence of leukemia in the laboratory bioassays showing formaldehyde to be carcinogenic. The results for benzene show that the paradigmatic human leukemogen doesn’t cause leukemia in this test system, so the null findings for formaldehyde should be given no weight.