Analysis of variance models were used to analyze 1153 personal samples of respirable granite dust collected in the Vermont granite sheds in 1970 and 1976. The best fitting model described dust concentrations in terms of job, shed, season, survey year, and included interaction terms between shed, season, and survey. This model explained 46% of the total variability in dust levels, and was used to estimate median annual dust concentrations for every job and shed in the industry. The analysis of variance suggested that dust levels varied more between sheds than between jobs within a shed, and that winter dust levels were significantly higher than summer levels. Dust levels in the 1970s were estimated by the average of the seasonally adjusted levels in the two surveys. Dust estimates in each job and shed were combined with personal work histories, and three procedures for estimating each worker's lifetime exposure were evaluated. The cumulative exposure index with uniform weights for each dust year of exposure was selected as the most physiologically plausible measure of total dose.