Recent works have shown that deep learning models effectively assist in the task of image classification, which is often applied in industrial problems, e.g., defective PCB detection. However, training a model for defect detection requires lots of training data, but nowadays, it is challenging to obtain defective samples due to the production line is stable and mature. Therefore, we design two defective PCB image generation methods for different defect types. The proposed generation methods can produce more defective PCB images as much as we want, and experimental results show that our proposed methods can generate realistic defective images.