The feeding habitat selection of blue sheep (Pseudois nayaur) was studied by direct observation method in the Helan Mountains, China during winter (from Novembe r to December) and spring (from April to June) from 2003 to 2004. We established 25 line transects to collect information on feeding habitats used by blue sheep. Blue sheep in the study area preferred mountain savanna forests, a habitat dominated by Ulmus glaucescens, with medium tree density (<4 individuals / 400 m 2 ), moderate tree height (4-6 m), higher shrub density (>5 individuals / 100 m 2 ), higher shrub (>1.3 m), higher food abundance (>50 g), moderate distance to human disturbance (<500 m), and mild distance to bare rock (<2 m). Such habitats characterized by 12 ecological factors were preferred as feeding areas by blue sheep during winter. Similar to habitat selection by the species during winter, blue sheep also showed a preference for mountain savanna with tree dominated by Ulmus glaucescens and medium tree density (<4 individuals / 400 m 2 ) during spring. Nevertheless, blue sheep preferred medium tree height (<6 m), moderate tree density (5-10 individuals / 100 m 2 ), medium shrub height (1.3-1.7 m), higher food abundance (>100 g), moderate altitude (<2 000 m), moderate distance to water resource (<500 m), and medium hiding cover (50%-75%) during spring. Selection of the feeding habitats by sheep showed a significant difference in vegetation type, landform feature, dominant tree, tree height, shrub density, distance to the nearest shrub, food abundance, slope direction, slope degree, distance to water resource, and hiding cover between winter and spring. Results of principal components analysis indicated that the first principal component accounted for 24.493% of the total variance among feeding habitat variance during winter, with higher loadings for vegetation type, dominant tree, tree height, distance to the nearest tree, shrub density, shrub height, altitude, distance to water resource, and distance to human disturbance. In spring, the first principal components explained 28.777% of the variance, with higher loadings for vegetation type, distance to the nearest tree, shrub height, distance to the nearest shrub, food abundance, altitude, and distance to human disturbance.