This project proposes a novel approach for detecting deepfake images utilising the MesoNet algorithm. MesoNet, a lightweight convolutional neural network (CNN), demonstrates exceptional efficiency in discerning subtle artefacts indicative of deepfake manipulation. The proposed methodology involves preprocessing the input images to extract discriminative features, which are then fed into the MesoNet architecture for classification. Leveraging transfer learning, the MesoNet model is fine-tuned on a curated dataset comprising both authentic and deepfake images, facilitating the learning of intricate patterns characteristic of synthetic alterations. Through extensive experimentation and evaluation on benchmark datasets, the efficacy of the proposed approach is thoroughly assessed, demonstrating superior performance in accurately identifying deepfake images while minimising false positives.