In the welfare of the elderly, it is important to detect the signs of dementia at an early stage and prevent it from becoming serious. We evaluated the performance of SVM-based cognitive function classification models and investigated the drawing features that contribute to distinguishing the severity of cognitive functions. Clock drawing test (CDT) was conducted on three groups of elderly people with different degrees of cognitive impairment. Feature selection was applied to the qualitative drawing features of the CDT, and a two-class classification model was constructed using support vector machine. The results showed that the five features related to conceptual deficits and spatial and planning deficits could be used to classify the dementia group and healthy control group with 79 % accuracy, and all the features showed statistically significant differences. It is suggested that these qualitative drawing features of the CDT can be applied to dementia screening.