Inspecting what sort of starch in commercial starchnoodles is important to international trade, food safety and protecting consumer benefit. At present, the inspection of components of starches in starch-noodle mainly relies on sensory perception, and which is fallibility or trustless. Because the microstructure pattern of starches in starchnoodles depends mainly on a kind or blend of starches from which the starch-noodle was made, this paper presents an approach to classify the starch-noodles by using computer system automatically based on recognizing the microstructure pattern of the starches and components in starch-noodle. The method consists of three step: 1) take the micrograph of starch-noodles with scanning electron microscopy and preprocessing. 2) extract features of fractal geometry and Gray-Level Co-Occurrence from micrograph. 3) distinguish a sort of starch-noodles by using these combined features as input vector of artificial neural networks. The experiments has been conducted with starch-noodles of mungbean blending pachyrhizus, and the experimental results show that the method is practicable and effective.