The current blind image SR reconstruction methods based on kernel estimation face a big challenge of accurately estimating blur kernels, while those based on component learning result in a complex network with a great number of parameters and a long inference time. To address these issues, this paper proposes a novel lightweight network, two-level edge-enhanced blind image super-resolution network (TLEESR), which consists of two key blocks: the edge-enhanced channel attention (EECA) block and the edge-enhanced information distillation (EEID) block. The EECA is designed based on complement learning to enrich the coarse-grained contour edge at the channel level by utilizing a cross-attention map between the edge map and the low-resolution (LR) feature map. The EEID further enhances fine-grained edge details through information distillation mechanism at the spatial level. In EEID, multi-branch convolution block is designed to extract much richer edge information during training stage, and simplified as a 3 × 3 convolution by structural re-parameters method during inference stage. Simulation results show that the proposed TLEESR achieves almost the same blind image reconstruction quality comparing to the current state-of-the-art blind SR methods (such as DANv3, DCLS, CDCN) on the × 4 SR task, and provides superior results on real blind SR image reconstruction, while remains a lightweight network with the fewest parameters and shortest inference time.