The Generative Adversarial Network (GAN) has made significant advances in the field of deep learning, compared to other generative models. GANs have the capability to generate images that look extremely realistic, which can be exploited for deceptive purposes, such as doctoring images or videos to spread false information on social media. As GANs continue to advance, the similarity between generated and real images becomes increasingly challenging to discern. Consequently, the need to distinguish between fake images generated by GANs and real ones becomes paramount. In this paper, we suggested a new approach for the detection of fake images generated by GANs. We introduced ELA-based deep learning techniques to detect real and fake images. The images from the dataset are preprocessed using Error level analysis (ELA) to check the image compression ratio and then passed to a pre-trained convolutional neural network model (ResNet50). Based on the transfer learning principle, the obtained features will be classified using three classifiers (Softmax, SVM, and Random Forest). Two datasets of fake images generated by GANs were used to evaluate the suggested method (Real and Fake Face detection dataset and CIFAKE dataset). On both datasets, excellent results are reached. The developed approach has shown exceptional performance in the detection of fake images generated by GANs.