- Spatial capture-recapture (SCR) is employed for estimating abundance and density of species, particularly those that are cryptic or solitary, and evaluating how population density varies with habitat. However, it is uncertain whether estimates are biased when applied to species that aggregate, such as elk (Cervus canadensis). Wildlife managers in the Pacific Northwest lack a reliable method to estimate abundance and density of Roosevelt elk (C. c. roosevelti), as that subspecies frequents dense forests and occurs singly or in groups sometimes exceeding 100 individuals. Hitherto, decision-making in elk management has relied on visual counts as population indices, yet such counts are potentially biased because group size influences detection and observers are unlikely to detect individuals in forests. We employed non-invasive sampling and spatial capture-recapture (SCR) modeling to estimate Roosevelt elk population density in two Oregon wildlife management units (WMUs), Tioga and McKenzie, and examined how density varied with habitat or land ownership type. We imposed a grid across both WMUs, basing cell size on elk home ranges that we estimated in these habitats from existing telemetry data, and stratified sampling by land ownership and nutrition quality, calculated from U.S. Forest Service Westside Elk Nutrition models. We randomly selected cells in each stratum and placed three, 2-km transects within each to facilitate recaptures. We sampled transects once in March to June 2018 and again in 2019 by walking a pre-determined bearing and searching for elk feces, tracking distance to control for variation in effort. We genotyped samples at 9 microsatellite loci and one sex-determining marker to identify individuals, then created a capture history for each individual. We evaluated a suite of SCR models to assess the effects of covariates relating to habitat type, terrain, precipitation, human activity, and sampling effort on elk population density and probability of detection. We applied the models to the capture histories, tested for the influence of aggregation on density estimates, and estimated mean elk population density for 2018 and 2019 at 0.80 and 0.20 individuals/km2 in Tioga and McKenzie, respectively. Our models performed well in areas with high elk density, deviating from true density 16% of the time. In contrast, our models deviated from true density 54% of the time when applied to areas with low elk density, indicating sampling intensity would need to be increased to obtain adequate recaptures. We did not find evidence that aggregation of individuals influenced our estimates of density in this system. Our results indicate that effort and precipitation influenced the probability of detecting an individual, and distance to forage/cover edge, distance to roads, percent slope, distance to crops, and precipitation influenced our estimates of elk population density. Overall, our models predicted fewer elk on federal lands, indicating that public recreational opportunities involving elk such as hunting and wildlife viewing may be more limited on public lands. Our methodology provides a framework for managers to develop and implement surveys to reliably estimate elk density in forested landscapes.