Few data are available about how social networks reduce the risk of cognitive impairment in old age. We aimed to measure this effect using data from a large, longitudinal, epidemiological clinicopathological study. 89 elderly people without known dementia participating in the Rush Memory and Aging Project underwent annual clinical evaluation. Brain autopsy was done at the time of death. Social network data were obtained by structured interview. Cognitive function tests were Z scored and averaged to yield a global and specific measure of cognitive function. Alzheimer's disease pathology was quantified as a global measure based on modified Bielschowsky silver stain. Amyloid load and the density of paired helical filament tau tangles were also quantified with antibody-specific immunostains. We used linear regression to examine the relation of disease pathology scores and social networks to level of cognitive function. Cognitive function was inversely related to all measures of disease pathology, indicating lower function at more severe levels of pathology. Social network size modified the association between pathology and cognitive function (parameter estimate 0.097, SE 0.039, p=0.016, R(2)=0.295). Even at more severe levels of global disease pathology, cognitive function remained higher for participants with larger network sizes. A similar modifying association was observed with tangles (parameter estimate 0.011, SE 0.003, p=0.001, R(2)=0.454). These modifying effects were most pronounced for semantic memory and working memory. Amyloid load did not modify the relation between pathology and network size. The results were unchanged after controlling for cognitive, physical, and social activities, depressive symptoms, or number of chronic diseases. These findings suggest that social networks modify the relation of some measures of Alzheimer's disease pathology to level of cognitive function.