Size distributions of airborne fiber exposures should be characterized for studies of respiratory disease because size determines the region of the lung where a fiber will deposit and its ability to produce toxic effects in cells. Yet fiber size is not measured precisely with standard air sampling methods. Specific fiber dimensions hypothesized to have biologic activity have been proposed, but these have not been evaluated in epidemiologic studies because there has not been a way to account for fiber size in historical air monitoring data. In this study, methods were developed to predict fibrous aerosol size fractions generated during glass wool fiber production using regression models and factors related to bulk fiber products and processing. A set of air samples representing a range of products and process applications was collected in eight fiber glass production facilities. The samples were analyzed more intensively than standard methods require. For each air sample, total fiber size distributions were measured using electron microscopy and two proportions were then calculated: (1) fibers meeting the size criteria of the standard NIOSH Method 7400 B rules method (pB), and (2) fibers meeting the size criteria for a biologically based exposure index, the hypothetically active fiber (HAF1) index (pH1). The fiber size proportions were used as dependent variables in regression models with production process factors. It was found that two factors, the nominal diameter of the bulk fiber product and whether oil was applied to it, determine more than 80% of the variability in the proportions (for the pB model, R2 = 0.86; for the pH1 model, R2 = 0.82). Using these two predicted proportions, it is possible to estimate the concentration of fibers in the biologically based HAF1 size fraction from a standard fiber concentration measurement. The models developed here can be used to size-adjust historical fiber concentration measurements for use in epidemiologic studies of respiratory disease.