Myelin water fraction (MWF) mapping can substantially improve our understanding of several demyelinating diseases. While MWF maps can be obtained from multi-exponential fitting of multi-echo imaging data, current solutions are often very sensitive to noise and modeling errors. This work addresses this problem using a new model-based method. This method has two key novel features: a) an improved signal model capable of compensating practical signal errors, and b) incorporation of parameter distributions and low-rank signal structures. Both simulation and experimental results show that the proposed method significantly outperforms the conventional methods currently used for MWF estimation.