The proliferation of fake images on the internet has become increasingly alarming. Advanced techniques including generative adversarial networks can generate visually real images that can mislead people and create false information. This poses a threat and can cause serious impacts. Many methods based on deep learning were proposed to detect fake images. These methods have demonstrated ability to achieve highly accurate results in detecting fake images. However, due to its “black-box” nature, there is a lack of explainability of the decision-making process in these models. In this paper, we integrate the convolution block attention module in ResNet-18 to improve the explainability of the deep learning model for fake image detection. The results showed that our method achieved a higher performance compared to the baseline method.