The San Francisco pedestrian volume modeling process refined the methodology used to develop previous intersection-based models and incorporated variables that were tailored to estimate walking activity in the local urban context. The methodology included two main steps. First, manual and automated pedestrian counts were taken at a sample of 50 study intersections with a variety of characteristics. A series of factor adjustments were applied to produce an annual pedestrian crossing estimate at each intersection. Second, log-linear regression modeling was used to identify statistically-significant relationships between the annual pedestrian volume estimate and land use, transportation system, local environment, and socioeconomic characteristics near each intersection. Twelve alternative models were considered, and the preferred model had a good overall fit (adjusted-R 2= 0.804). As identified in other communities, pedestrian volumes were positively associated with the number of households and the number of jobs near each intersection. Uniquely, this San Francisco model also found significantly higher pedestrian volumes at intersections in high-activity zones with metered on-street parking, in areas with fewer hills, near university campuses, and controlled by traffic signals. The model was based on a relatively small sample of intersections, so the number of significant factors was limited to six. Results are being used by public agencies in San Francisco to better understand pedestrian crossing risk and to inform citywide pedestrian safety policy and investment.