Convolutional Neural Networks (CNNs), inspired by the local receptive fields of the visual cortex in the brain, have been widely used across various domains, including image recognition. The unraveling of the mechanisms of visual illusion generation via CNNs could enhance our understanding of the brain’s visual processing mechanisms and potentially aid in recognizing and mitigating the risks of misclassification by CNNs. In this study, we used spiral illusion images to verify whether the illusion occurs in a CNN. We constructed a CNN model and trained it to distinguish between concentric circle images and spiral images. We then introduced 14 spiral illusion images into the trained CNN model and tasked it with classifying them as either concentric circle images or spiral images. This process, from training to classification, was repeated 10 times, and the results were aggregated. The findings revealed that in all trials, 57% of the images (8 out of 14) were classified as spiral images, indicating the occurrence of a spiral illusion. Conversely, 29% of the images (4 out of 14) were consistently classified as concentric circle images, suggesting the absence of a visual illusion. These results suggest that the CNN model developed in this study is highly likely to generate spiral illusions. Furthermore, it was deduced that certain images are more prone to inducing illusions than others. Specifically, images with thick black-and-white lines alternately connected at the ends of concentric circles were the most prone to generate the spiral illusion.