We assessed the relative importance of various factors contributing to spawning site use by a population of threatened coho salmon Oncorhynchus kisutch in Freshwater Creek, California, and created a predictive model of spawning habitat selection based on logistic regression analysis. We excluded sampling sites that previous studies had established as unsuitable on the basis of depth and substrate criteria and asked why fish chose particular locations and not others in seemingly suitable habitat. We evaluated surface water velocity, depth, substrate size composition, gravel inflow rates, vertical hydraulic gradient, geomorphic channel units, hyporheic water physicochemistry, cover, and proximity to other redds not in sampling sites during the 2004-2005 spawning season. In univariate comparisons with unused sites, coho salmon selected sites with a smaller median particle diameter, a larger percentage of gravel-pebble substrate, and higher gravel inflow rates. Based on multivariate logistic regression, the probability of a site's being used for spawning was best modeled as a positive function of the gravel-pebble fraction of the substrate, location at a pool or run tail, and the presence of existing redds in close proximity to the site. This model explained 38% of the variation in the data and was a better predictor of spawning habitat use than a more traditional model based on depth, velocity, and substrate. Our results highlight the potential importance of social behavior in contributing to habitat selection by spawning salmonids. ?? Copyright by the American Fisheries Society 2007.