In this study, we developed a hand hygiene assessment system using pix2pix, a type of generative adversarial network, by imaging the palms of nursing students after handwashing. For this, a fluorescent lotion was used for hand-wash training, and a black light was used to visualize the remaining residues after washing. In pix2pix, the adopted input image was a black light image obtained after handwashing, while the ground truth image was a binarized image created by extracting the residues remaining in the input image, as determined by a trained staff member. We used 433 paired images after handwashing as training data and employed 30 images for verification. To evaluate the training model, we used the intersection over union (IoU) metric to assess the degree of overlap between the generated image and the ground truth image. The results indicated that the IoU value was highest when the training model with data expansion was used (p < 0.05). Subsequently, the effectiveness of this assessment method was verified using data from six handwashing measurements performed on the nursing students each week. The results showed that the percentage of unwashed hands decreased significantly after the fourth measurement (p < 0.05).