Although current deep learning-based methods have gained promisingperformance in the blind single image super-resolution (SISR) task, most ofthem mainly focus on heuristically constructing diverse network architecturesand put less emphasis on the explicit embedding of the physical generationmechanism between blur kernels and high-resolution (HR) images. To alleviatethis issue, we propose a model-driven deep neural network, called KXNet, forblind SISR. Specifically, to solve the classical SISR model, we propose asimple-yet-effective iterative algorithm. Then by unfolding the involvediterative steps into the corresponding network module, we naturally constructthe KXNet. The main specificity of the proposed KXNet is that the entirelearning process is fully and explicitly integrated with the inherent physicalmechanism underlying this SISR task. Thus, the learned blur kernel has clearphysical patterns and the mutually iterative process between blur kernel and HRimage can soundly guide the KXNet to be evolved in the right direction.Extensive experiments on synthetic and real data finely demonstrate thesuperior accuracy and generality of our method beyond the currentrepresentative state-of-the-art blind SISR methods. Code is available at:https://github.com/jiahong-fu/KXNet.