Abstract: According to cotton-flax blended fabric, a new automatic identification method of microscopic fiber images in the longitudinal view in which the length of fiber is about 0.5 mm is introduced in this paper. In fiber detection section, the background of fiber image is removed first, then fiber areas are detected by a method combining morphological close computing and background regional growth method, which filters glass scratches and other sundries well. Based on the region image, binary image and thinning image of binary image in fiber skeleton vertical direction and their vertical integral projections, the coefficient variation (CV) and mean for each of the projections are obtained. Such six statistics parameters are used as the characteristic parameters for identification of cotton fibers and flax fibers. The training set is used to train the least square support vector machine classifier. The experiments for the testing set showed that the mean identification accuracy of cotton and flax fiber is 93.3%.