Currently, artificial intelligence technology can create or modify audio, video, or image content to look and sound like the original, known as deepfake. Artists and content creators can use deepfake to create realistic visual effects in movies or games. However, Deepfake has great potential to be used as a tool for spreading false information with the ability to create videos, images and audio that look very authentic, this technology can be used to manipulate reality and spread fake news or slander. The uncontrolled spread of deepfake can reduce the level of public trust in the media and information sources, creating significant uncertainty and social instability. One effort to overcome this problem is to develop a system that can detect deepfake. This research aims to develop a system that can detect deepfake using the MesoNet algorithm. MesoNet is part of Convolutional Neural Network (CNN) which consists of a neural network with few neural layers and concentrates on mesoscopic features of images, which include low-level and high-level features. The deepfake detection model that has been created by applying the MesoNet algorithm shows high performance values, namely with an accuracy value of 90.93%, a precision value of 87.78%, a recall value of 95.09% and an f1-score value of 91.28%.