Because it has similar spectral characteristics as glaciers, snow around glaciers is a complicating factor in glacier inventory from satellite images. It is rare to acquire snowless and cloud-free satellite images in perennial snow mountain regions, and, moreover, glaciers are also usually shaded by mountain shadows. Therefore, a monotemporal satellite image can hardly map an intact glacier boundary. An object-oriented image segmentation method is proposed to map glaciers and their changes with multitemporal Landsat imagery. First, a “global-local” iteration segmentation method was used to delineate the snow and ice boundaries with each single phase image. Second, the mountain shadows of multitemporal Landsat images with different sun angles were mapped by methods of terrain analyses, and some parts of the glaciers inside these shadow regions can be identified. Finally, the complete glacier boundaries were restored with multitemporal images, and then these boundaries were intersected to get the minimum glacier extents. The method was tested with multitemporal Landsat images during 1977–2013 in the eastern Tienshan Mountains of Central Asia, and the spatial patterns and temporal process of these glacier changes in the last years were also analyzed. The results showed that the proposed method performs well in mapping glaciers in rugged alpine regions. Combining multiple images in different seasons provides a more effective way of removing disturbing factors during the process of glacier extraction. The total glacier area in the eastern Tienshan Mountains has decreased to 106.83 km2 in 2013, with a loss of 21.5% since 1977, and the rate of glacier shrinkage accelerates in recent decades. The relationship between glacier recession and river runoffs was also analyzed, and the results showed that small glaciers whose areas are about 1.1 km2 provide less runoff than those larger glaciers with about 2.5 km2.