We present a novel loss function, namely, GO loss, for classi cation. Most of the existing methods, such as center loss and contrastive loss, dynamically determine the convergence direction of the sample features during the training process. By contrast, GO loss decomposes the convergence direction into two mutually orthogonal components, namely, tangential and radial directions, and conducts optimization on them separately. e two components theoretically a ect the interclass separation and the intraclass compactness of the distribution of the sample features, respectively. us, separately minimizing losses on them can avoid the e ects of their optimization. Accordingly, a stable convergence center can be obtained for each of them. Moreover, we assume that the two components follow Gaussian distribution, which is proved as an e ective way to accurately model training features for improving the classi cation e ects. Experiments on multiple classi cation benchmarks, such as MNIST, CIFAR, and ImageNet, demonstrate the e ectiveness of GO loss.