We conducted a cross-sectional epidemiologic study to assess the association between blood lead level and hematocrit in 579 one to five year-old children living near a primary lead smelter in 1974. Blood lead levels ranged from 0.53 to 7.91 mumol/L (11 to 164 micrograms/dl). To predict hematocrit as a function of blood lead level and age, we derived non-linear regression models and fit percentile curves. We used logistic regression to predict the probability of hematocrit values less than 35 per cent. We found a strong non-linear, dose-response relationship between blood lead level and hematocrit. This relationship was influenced by age, but (in this age group) not by sex; the effect was strongest in youngest children. In one year-olds, the age group most severely affected, the risk of an hematocrit value below 35 percent was 2 percent above background at blood lead levels between 0.97 and 1.88 mumol/L (20 and 39 micrograms/dl), 18 percent above background at lead levels of 1.93 to 2.85 mumol/L (40 to 59 micrograms/dl), and 40 percent above background at lead levels of 2.9 mumol/L (60 micrograms/dl) and greater; background was defined as a blood lead level below 1.88 mumol/L (20 micrograms/dl). This effect appeared independent of iron deficiency. These findings suggest that blood lead levels close to the currently recommended limit value of 1.21 mumol/L (25 micrograms/dl) are associated with dose-related depression of hematocrit in young children.