Infrared images have the disadvantages of narrow dynamic range, low contrast and being easy to be polluted by noise. However, traditional infrared image denoising algorithms may filter out image details while removing noise. A new infrared image denoising method based on sparse representation is proposed. The method firstly clusters an original infrared image; secondly decomposes each cluster sub-image into a dictionary; and then the denoised infrared image is reconstructed from the sparse coefficient matrix. The experimental results show that this method can retain image details better than the traditional infrared image denoising algorithm and the visual effect is ideal.