A super-resolution method based on sparse dictionary is presented. The method efficiently builds sparse association between high-frequency components of HR image patches and LR image feature patches, and defines the association as a prior knowledge to guide super-resolution reconstruction based on sparse dictionary. Compared with overcomplete dictionary, sparse dictionary is more compact and effective to express the prior knowledge. We choose the high-frequency component of the HR image patch as its feature for dictionary training, which builds the sparse association between LR image patches and HR ones with better efficiency and less training examples. Sparse K-SVD algorithm is adopted as optimization method to improve the computation efficiency. Experiments with natural images show that our method outperforms several other learning-based super-resolution algorithms.