In this paper, we propose an evaluation method of riding comfortability for a wheel-chair on roads by using deep learning with the color spectrogram images which is creat-ed from acceleration data. First, we attach an acceleration sensor to the wheelchair and make the color-spectrogram images, which show the short-time changes of the frequen-cy of vibrations, from the obtained acceleration signals. Next, using the spectrogram im-ages, we make an evaluation model for wheelchair vibration classification by deep learn-ing. In the modeling, we performed a questionnaire survey to the persons who are passed to roads by the wheelchair, and asked them the classification of their riding comfortability as a wheelchair user, and created a classification model based on the evaluation criteria from the questionnaire. Finally, we perform the evaluation experi-ments with the wheelchair on external roads and evaluate the comfortabilities at each point on the experiment roads.