In this paper an algorithm for automatic segmentation of neonatal brain MRI is proposed. The first step of the process involves skull stripping and then noise removal using anisotropic diffusion filter. The image is then segmented into three clusters namely white matter, grey matter and cerebrospinal fluid. The segmentation strategy is based on expectation maximization of Gaussian mixture model of tissues. Each pixel is assigned to the tissue cluster which showed the maximum probability of being in. The pixels with same probability likelihoods for more than one cluster are also dealt successfully. The results are evaluated using Dice similarity coefficient. A maximum Dice similarity coefficient of 0.832 is obtained for white matter on T2 images. © 2016 IEEE.