Durian is a tropical plant native to Southeast Asia known as the famous fruit. The leaves that have their respective characteristics are different from the others. In previous studies, the percepton method (MLP) was used to identify the type of durian leaf. The study used the Convolutional Neural Network (CNN) method to classify durian leaf varieties. The research phases include preparation, library study, planning, data collection, implementation, and testing. The data collection involved images of durian leaves from the garden, and has four types of dorian leaves. The training, accuracy, and confusion matrix graphs showed a good model performance, especially in the 10th epoch. The model evaluation results showed an 80%, with adequate precision, recall, and F1 scores. The prediction table displays the prediction results for data samples, especially in the Old Monthong class, with a high degree of confidence (91.55%).