We described and predicted spatial variation in marine migration (anadromy) of female Oncorhynchus mykiss in the John Day River watershed, Oregon. We collected 149 juvenile O. mykiss across 72 sites and identified locations used by anadromous females by assigning maternal origin (anadromous versus non-anadromous) to each juvenile. These assignments used comparisons of strontium to calcium ratios in otolith primordia and freshwater growth regions to indicate maternal origin. We used logistic regression to predict probability of anadromy in relation to mean annual stream runoff using data from a subset of individuals. This model correctly predicted anadromy in a second sample of individuals with a moderate level of accuracy (e.g., 68% correctly predicted with a 0.5 classification threshold). Residuals from the models were not spatially autocorrelated, suggesting that remaining variability in the expression of anadromy was due to localized influences, as opposed to broad-scale gradients unrelated to mean annual stream runoff. These results are important for the management of O. mykiss because anadromous individuals (steelhead) within the John Day River watershed are listed as a threatened species, and it is difficult to discern juvenile steelhead from non-anadromous individuals (rainbow trout) in the field. Our results provide a broad-scale description and prediction of locations supporting anadromy, and new insight for habitat restoration, monitoring, and research to better manage and understand the expression of anadromy in O. mykiss.