This paper deals with the mensuration of super-micronized granule using computer vision. Image constrast was strengthened by non-linearity transformation, and target was segmented from background using auto-threshold algorithm. A recognition algorithm was designed to remove agglomerate granule in order to get real granule sizes and layout of samples. Experiments prove that this method of micro-image inspection of granule sizes can determine layout status directly perceived through the senses. The device can detect granule size from 0.1 μm to 150 μm and the result of inspection is stable.