Keen's myotis (Myotis keenii) has one of the most limited geographic distributions of any species of bat in North America. Because there is little knowledge of its roosting ecology, we examined selection of day-roosts in trees by male and female Keen's myotis at 3 spatial scales (tree, tree plot, and landscape) on Prince of Wales Island, southeastern Alaska, from May to September 2006. We selected variables known to influence roost selection by other tree-roosting bats for logistic regression models. We used Akaike's information criterion to rank all models within and between scales according to their ability to differentiate between characteristics of used and available roosts and we determined the effect of each variable with model-averaged coefficient estimates and associated odds ratios. We tracked 13 females and 6 males to 62 and 24 roosts in trees, respectively. Selection of day-roosts by males and females was most strongly influenced by characteristics of trees. The odds a tree was used for roosting by female Keen's myotis increased with the presence of defects, increasing diameter, and decreasing bark; increasing quadratic mean diameter in the tree plot; and decreasing distance to the nearest stream and increasing proportion of old growth in the landscape. Male Keen's myotis exhibited flexibility in types of roosts chosen, but the odds of a tree being used increased with decreasing bark, the presence of defects, and increasing slope-height. The odds a tree was used as a roost by males also increased with the increasing proportion of trees in early to late decay stages in the tree plot. Some habitat features differed between males and females at each spatial scale and differences are likely a reflection of the energetic demands associated with reproduction. We suggest that maintaining structural components characteristic of old-growth rain forest will promote conservation of Keen's myotis in southeastern Alaska.