The paper introduces MemNet, a very deep persistent memory network designed to address the long-term dependency problem in image restoration tasks. MemNet incorporates a memory block that includes a recursive unit and a gate unit, enabling the network to learn and retain persistent memory through an adaptive learning process. The recursive unit generates multi-level representations of the current state under different receptive fields, while the gate unit controls the retention of previous states and the storage of new states. MemNet is applied to three image restoration tasks: image denoising, super-resolution, and JPEG deblocking. Comprehensive experiments demonstrate the effectiveness of MemNet, showing its superior performance over state-of-the-art methods in all three tasks. The code for MemNet is available at https://github.com/tyshiwo/MemNet.The paper introduces MemNet, a very deep persistent memory network designed to address the long-term dependency problem in image restoration tasks. MemNet incorporates a memory block that includes a recursive unit and a gate unit, enabling the network to learn and retain persistent memory through an adaptive learning process. The recursive unit generates multi-level representations of the current state under different receptive fields, while the gate unit controls the retention of previous states and the storage of new states. MemNet is applied to three image restoration tasks: image denoising, super-resolution, and JPEG deblocking. Comprehensive experiments demonstrate the effectiveness of MemNet, showing its superior performance over state-of-the-art methods in all three tasks. The code for MemNet is available at https://github.com/tyshiwo/MemNet.