Winter bedding sites used by the roe deer Capreolus capreolus were studied using a resource selection function in the wetland of Sanjinag National Nature Reserve, Heilongjiang Province, from January to March 1999. Site characteristics were measured at the bedding sites and control sites. Roe deer selected diurnal bedding sites with island forest and clumps of reeds, denser canopy closure (>10%), higher numbers of new shoots (>30 individuals), moderate hiding cover (90%-99%), closer to water resources (<400 m), closer to croplands (<500 m), higher temperate index (>0.2 ℃), deeper snow (>31 cm) and farther from human disturbance (>1 000 m). At night, roe deer preferred bedding below island forests, denser canopy closure (>10%), higher numbers of new shoots (>30 individuals), mild hiding cover (<90%), moderate distance from water resources (<700 m), farther from croplands (>500 m), higher temperate index (>0.2 ℃) and farther from human disturbance (>1 000 m). Roe deer showed significant differences in use of scraped beds by deer, distance from cropland, snow depth, distance from human disturbance and width of beds between day and night. Day beds were associated with closer proximity to cropland, deeper snow and were narrower than night beds. The resource selection function of day beds used by roe deer is a logistical regression model: logit(P)=-13.999-0.446×vegetation type-0.003×distance from water resource-0.003×distance from cropland+0.915×temperate index+0.487×snow depth+0.001×distance from human disturbance. The selection probability is therefore: P=e logit(P)／1+e logit(P) . The model correctly classified 90.9% of day bed sites. The resource selection function of night beds by roe deer is: log it(P)=-0.009-1.863×vegetation type-0.004×distance from water resource+0.946×temperate index+0.002×distance from human disturbance. We can estimate the selection probability from this model: P=elogit(P)／1+e logit(P) . This model correctly classified 91.8% of night bed sites.