This work proposes a 3D framework for convolutional sparse coding of spectral images using dictionary learning. This scheme was tested within a denoising scheme, improving the state-of-the-art's representation approach by up to 5dB of PSNR.