Radiotherapy is a common treatment for tumors. To accurately control the radiation dose distribution and reduce the damage caused by radiation to normal tissues and organs in radiotherapy, organs at risk must be delineated precisely. In this study, a novel automatic segmentation method for organs in head and neck, named SAU-Net, was proposed, the architecture of which is based on the three-dimensional (3D) U-Net with residual connections. Nonlocal spatial attention implemented using the squeeze and attention (SA) module was introduced to solve the problem of unbalanced segmentation accuracy caused by massive differences in organs' volumes. This introduction increased the ability to aggregate multiscale contextual information by encoding global features. To avoid the stacking of excess local information by extra convolution operations and reduce the number of parameters, the model reduced the number of convolution kernels. The performance of the model was evaluated using the dice score, and SAU-Net achieved 13.7% and 8.2% higher segmentation accuracy than 3D U-Net and 3D residual U-Net (ResU-Net), respectively. Moreover, the proposed model achieved of an inference time 73% faster than that achieved by FocusNetv2. Thus, SAU-Net delineates organs at risk in the head and neck faster than AnatomyNet and more accurately than FocusNetv2.