In recent years, “kawaii” has been attracting attention as an affective value in manufacturing for various purposes. One example is the use of “kawaii characters” in marketing, PR and advertisement for several target groups especially young people. However, those “kawaii characters” have been designed and used intuitively without systematically evaluating their kawaii degree for target group. Since different target groups might have different preferences for kawaii characters, only intuitive design of kawaii characters might not fulfil their satisfaction and attract enough attention as expected. Therefore, this study proposes a systematic method to evaluate kawaii characters by constructing a model to classify kawaii characters with different physical attributes. To construct “kawaii character” dataset, we firstly prepared ten standard characters as images. Then, for each standard character, we prepared four different variations for each of these six physical attributes: eyebrows, eyes, mouth, facial (cheek) redness, clothing, and hair accessories. Next, we conducted a questionnaire to evaluate the kawaii degree of each kawaii character, and calculated it as “kawaii score”. Using the questionnaire results, we built a dataset containing a total of 120 images of kawaii characters and their corresponding kawaii scores. The dataset was used to construct a model using Deep Convolutional Neural Network (CNN) algorithm, which is a binary classification of kawaii characters into “kawaii” and “not-kawaii” group. Finally, we evaluated the classification performance of the model to confirm its performance for evaluating kawaii characters.