The appropriate exposure metrics for characterizing manganese (Mn) exposure associated with neurobehavioral effects have not been established. Blood levels of Mn (B-Mn) provide a potentially important intermediate marker of Mn airborne exposures. Using data from a study of a population of silicon- and ferro-manganese alloy production workers employed between 1973 and 1991, B-Mn levels were modeled in relation to prior Mn exposure using detailed work histories and estimated respirable Mn concentrations from air-sampling records. Despite wide variation in exposure levels estimated for individual jobs, duration of employment (exposure) was itself a strong predictor of B-Mn levels and strongest when an 80-day half-life was applied to contributions over time (t = 6.95, 7.44, respectively; p < 10-5). Partitioning exposure concentrations based on process origin into two categories: (1) "large" respirable particulate (Mn-LRP) derived mainly from mechanically generated dust, and (2) "small" respirable particulate (Mn-SRP) primarily electric furnace condensation fume, revealed that B-Mn levels largely track the small, fume exposures. With a half-life of 65 days applied in a model with cumulative exposure terms for both Mn-LRP (t = -0.16, p = 0.87) and Mn-SRP (t = 6.45, p < 10-5), the contribution of the large-size fraction contribution was negligible. Constructing metrics based on the square root of SRP exposure concentrations produced a better model fit (t = 7.87 vs. 7.44, R2 = 0.2333 vs. 0.2157). In a model containing both duration (t = 0.79, p = 0.43) and (square root) fume (t = 2.47, p = 0.01) metrics, the duration term was a weak contributor. Furnace-derived, small respirable Mn particulate appears to be the primary contributor to B-Mn levels, with a dose-rate dependence in a population chronically exposed to Mn, with air-concentrations declining in recent years. These observations may reflect the presence of homeostatic control of Mn levels in the blood and other body tissues and be useful in assessing Mn exposures for evaluating neurotoxic effects. Corrigendum.