In this paper, a novel multi-scale method for coil sensitivity profile correction is presented based on wavelet transform. A magnetic resonance (MR) image can be decomposed into two spaces by the wavelet transform: approximate space and residual space. The approximate templates in approximate space can be thought of as multi-scale sensitivity profiles of the surface coil for coil correction. When we choose a suitable filter for decomposition, one of the sensitivity profiles should be optimal among the multi-scale sensitivity profiles. The optimal sensitivity profile can be chosen automatically by an analysis of the contents of the two spaces. The multi-scale method does not rely on any data other than the image generated by the MR scanner. The experiment showed promising results based on one-dimensional simulation and images of phantom and human images.