Abstract Background There is international consensus that benzene exposure is causally related to acute myeloid leukemia (AML), and more recent evidence of association with myelodysplastic syndromes (MDS). However, there are uncertainties about the exposure response, particularly risks by time since exposure and age at exposure. Methods In a case–cohort study in 110 631 Chinese workers followed up during 1972–1999 we evaluated combined MDS/AML (n = 44) and chronic myeloid leukemia (n = 18). We estimated benzene exposures using hierarchical modeling of occupational factors calibrated with historical routine measurements, and evaluated exposure response for cumulative exposure and average intensity using Cox regression; P values were two-sided. Results Increased MDS/AML risk with increasing cumulative exposure in our a priori defined time window (2 to <10 years) before the time at risk was suggested (Ptrend = 08). For first exposure (within the 2 to <10-year window) before age 30 years, the exposure response was stronger (P = .004) with rate ratios of 1.12 (95% confidence interval [CI] = 0.27 to 4.29), 5.58 (95% CI = 1.65 to 19.68), and 4.50 (95% CI = 1.22 to 16.68) for cumulative exposures of more than 0 to less than 40, 40 to less than 100, and at least 100 ppm-years, respectively, compared with no exposure. There was little evidence of exposure response after at least 10 years (Ptrend = .94), regardless of age at first exposure. Average intensity results were generally similar. The risk for chronic myeloid leukemia was increased in exposed vs unexposed workers, but appeared to increase and then decrease with increasing exposure. Conclusion For myeloid neoplasms, the strongest effects were apparent for MDS/AML arising within 10 years of benzene exposure and for first exposure in the 2 to less than 10-year window before age 30 years.