For granular formulations of pesticides, direct consumption by birds is generally the most important route of exposure. A probabilistic exposure model was developed that estimates how many pesticide granules a bird ingests and, from that, the quantity of pesticide ingested. This model, referred to as the "granular pesticide avian risk assessment model" (GranPARAM), has input variables not included in current screening-level assessments for granular pesticides, such as proportion of time for which birds forage in the field, grit ingestion rates, attractiveness of pesticide granules compared with natural grit, and proportions of soil particles and pesticide granules in the size range consumed by birds. For input variables that are uncertain, variable, or both, distributions are used rather than point estimates. Monte Carlo analysis is then performed to propagate input variable uncertainties through the exposure model for granular pesticides. The outputs from the exposure portion of GranPARAM are estimated pesticide doses for each of 20 birds of a selected species on each of 1000 fields. The dose for each bird is compared with a randomly chosen dose from the dose-response curve for that species or an appropriate surrogate. If the exposure dose for a bird exceeds the randomly chosen effects dose, the bird is considered dead; otherwise, the bird is assumed to be alive. Thus, the risk output from GranPARAM is a bar chart showing the percentages of fields with 0/20 dead birds, 1/20 dead birds, 2/20 dead birds, and so forth.