This paper proposes a Trainable Feature Matching Attention Network (TFMAN) for single image super-resolution (SISR). TFMAN integrates two key modules: Trainable Feature Matching (TFM) and Same-size-divided Region-level Non-Local (SRNL). TFM explicitly learns features from training data through feature matching, enhancing the representation ability of CNNs. SRNL reduces computational and memory demands by applying non-local operations in parallel on uniformly divided blocks of the input feature map. The network uses a recurrent convolutional network as its backbone and includes three branches for feature fusion. TFM is followed by a Channel Attention (CA) module to focus on important channels. Experiments on benchmark datasets show that TFMAN achieves superior performance with fewer parameters compared to state-of-the-art methods. The code is available at https://github.com/qizhou000/tfman. Keywords: Super-resolution, feature matching, non-local, recurrent convolutional neural network, deep learning.This paper proposes a Trainable Feature Matching Attention Network (TFMAN) for single image super-resolution (SISR). TFMAN integrates two key modules: Trainable Feature Matching (TFM) and Same-size-divided Region-level Non-Local (SRNL). TFM explicitly learns features from training data through feature matching, enhancing the representation ability of CNNs. SRNL reduces computational and memory demands by applying non-local operations in parallel on uniformly divided blocks of the input feature map. The network uses a recurrent convolutional network as its backbone and includes three branches for feature fusion. TFM is followed by a Channel Attention (CA) module to focus on important channels. Experiments on benchmark datasets show that TFMAN achieves superior performance with fewer parameters compared to state-of-the-art methods. The code is available at https://github.com/qizhou000/tfman. Keywords: Super-resolution, feature matching, non-local, recurrent convolutional neural network, deep learning.