This paper proposes a multiresolution Matching Pursuit decomposition of natural images. Matching Pursuit is a greedy algorithm that decomposes any signal into a linear expansion of waveforms taken from a redundant dictionary, by iteratively picking the waveform that best matches the input signal. Since the computational cost rapidly grows with the size of the signal, we propose a multiresolution strategy that, together with a dictionary training, significantly reduces the encoding complexity while still providing an efficient representation. Such a decomposition is perceptually very effective at low bit rate coding, thanks to similiarities with the Human Visual System information processing.