Tree rubbing or marking by bears has been observed throughout the northern hemisphere. Even so, this behaviour has rarely been studied. We documented 93 sites where grizzly bears Ursus arctos horribilis rubbed on 116 trees during 1986–1992, in the Yellowstone Ecosystem. We used logistic regression and information-based estimation and selection criteria to specify models that explained selection of sites and individual trees for rubbing by bears in our study area. The probability of rubbing peaked during May and June, the period of mating and moult, and declined thereafter. At the landscape level, grizzly bears selected for gentle south-facing slopes, forest/non-forest ecotones with sparse deadfall, and forest stands dominated by lodgepole pine Pinus contorta or Douglas-fir Pseudotsuga menziesii. Among the trees at sites where bears rubbed, we found strong selection for large diameters but no indication of selection for species. Rubbed trees were highly associated with travel routes likely used by bears, including game trails, recreation trails and forest edges. Rubbing was often oriented towards these likely travel routes. Short trails of entrenched pad-shaped marks leading up to rubbed trees were recorded at 58% of the sites where rubbing occurred. Contrary to reports of black bears Ursus americanus clawing and biting trees, we found shredded or bitten bark at only 9% of sites with rubbed or otherwise marked trees. Circumstantial evidence suggests that bears used trees primarily for rubbing their back and shoulders. Our findings are consistent with previous arguments that rubbing serves as a means of chemical communication.