We investigated the use of deep learning for shape classification of ash sampled from actual pulverized coal boilers to evaluate the ash deposition phenomenon. As a method, images of individually sorted ash particles were classified into seven different shapes and evaluated with each of the five ash samples. The result showed that the number and mass distributions of each shape were characterized differently with samples. Based on the mechanism of ash deposition on heat exchange tubes, “fine” and “agglomerated” particles were thought to be involved in ash deposition. It was considered that these particle shapes could be used as a new method to evaluate the ash deposition phenomenon.