There has been increasing interest in recent years in developing strategies in epidemiology for the summarization of occupational exposures, strategies that serve to clarify observed relationships between occupational exposure and health outcomes. Where source occupational exposure data are scarce, it is common to assemble exposure groups with the goal of increasing the extent to which data-based exposure estimates are available for an entire cohort. There has been little guidance, however, on the effect of different grouping strategies on the observed fit between exposure and health outcome. This investigation examined the effect of the use of different exposure summarization strategies on observed relationships between dust exposure and lung function decline among coal mine workers. The dust exposure and spirometry data employed were gathered in the National Study of Coalworkers' Pneumoconiosis. An analysis of variance procedure was carried out to characterize the variability of the dust exposure data, employing single variables relating to mine identity, occupation, and year, as well as two- and three-way multivariate combinations of these variables. The resulting combinations were ranked according to the standard deviation of the observed exposure range to reflect the relative specificity of the various approaches. Sequential arrangements of single- and multiple-variable combinations were constructed, alternately employing highly specific codes or broad categories for mine, occupation, and year. Annual exposure estimates were constructed on the basis of these sequences and used in tandem with longitudinal change in forced expiratory volume (FEV1) in linear multiple regression procedures. Height, age, smoking status, and dust exposure were employed as predictor variables. The results show that the use of broad categorization approaches had a substantial impact on observed regression coefficients. The largest change was observed for categorization according to occupation, which resulted in two- to three-fold increases in the magnitude of observed regression coefficients. These results suggest that the use of highly specific exposure summarization approaches may result in regression outcomes which are marked by a high degree of attenuation, and that consideration of the precision of summarized exposure estimates is an important component of an effective exposure assessment strategy.