In semiconductor manufacturing, wafer defect classification is an important step in ensuring product quality. However, due to the diversity and complexity of wafer defects, the existing hybrid wafer defect classification network still has shortcomings in accuracy. To solve this problem, a hybrid wafer defect classification network based on global and local multi-scale feature fusion—MLG-Net was proposed. MLG-Net consists of three main modules: feature extraction module, global branch, and local branch. The network aims to better extract and utilize the global semantic information and local detail features of wafer defect images, which are combined with multi-scale feature fusion technology to form a more comprehensive feature representation, which helps the classifier to make more accurate judgments in the face of complex mixed defects, thereby improving the classification accuracy. To verify the effectiveness of MLG-Net, a large number of experiments were carried out on MixedWM38, a dataset containing 38 mixed types of defects, and the classification accuracy reached 98.84%. The results show that MLG-Net is superior to the six mainstream wafer defect classification methods in terms of comprehensive performance. This result demonstrates the importance and effectiveness of global and local feature fusion in dealing with hybrid wafer defect classification tasks.