Abstract Natural regeneration in canopy gaps is a key process affecting long-term dynamics of many forests, including northern hardwood forests. The density and composition of regenerating trees are often highly variable, reflecting sensitivity to a suite of driving factors operating at different scales (e.g., harvest gap to regional landscape), including production of seeds, physical characteristics of gaps and stands, competition with non-tree vegetation, and browsing by animals. Multivariate analyses over broad geographic areas provide insights into the relative effects of these factors and permit exploration of spatial patterns in regeneration. We examined the effects of gap-, stand-, and landscape-scale factors on densities of tree seedlings ( 2 region of Michigan's Upper Peninsula. We used Bayesian multilevel modeling to account for the hierarchical structure of the data and assess uncertainty in parameter estimates. Sugar maple ( Acer saccharum ) saplings were absent from 61% of 154 m 2 plots centered in harvest gaps ( n = 347) despite its high shade tolerance and overstory dominance, but densities were high in other gaps. Densities of sugar maple seedlings and/or saplings were negatively associated with a combination of greater stand-scale densities of white-tailed deer ( Odocoileus virginianus ), greater gap-scale cover of non-tree vegetation, and lower gap-scale light availability, with deer density having the greatest effect. Densities of unpalatable and commercially less valuable ironwood ( Ostrya virginiana ), the second most common regeneration species, were positively related to gap-scale seed-production potential but were unrelated to factors affecting sugar maple. Ironwood tended to replace sugar maple saplings in areas with high deer density. At the landscape scale, densities of sugar maple seedlings and saplings decreased with decreasing latitude and snow depth and increasing winter deer densities. These inverse spatial patterns suggest that deer herbivory can lead to landscape-scale variation in regeneration success. However, the spatial distribution of habitat types (a proxy for soil moisture and nutrient conditions) confound this observation, with higher densities of sugar maple generally located on stands with less nutrient-rich habitat types. Results demonstrate that combinations of factors operating at different scales, and with different relative magnitudes of impact, contribute to high variation in regeneration composition and density following timber harvest. Selection silvicultural practices, as currently applied, do not ensure regeneration of desirable species; practices might require modifications in general (e.g., increasing gap size) and to match them to regionally varying factors like deer density.