A water referencing algorithm for addressing the spectroscopic imaging problems of low SNR and main field inhomogeneities is proposed. Using the location of the water peak from each voxel and additional a priori information results in a parametric estimation problem. Optimum estimates of the desired metabolite concentration can then be computed and displayed in an image format. The algorithm is shown to be very stable in the presence of noise and is insensitive to Bo inhomogeneity. A detailed error analysis as well as extensions to the basic data model are also discussed. Results from both 1H and 31P experiments are presented to verify the predicted good performance even with extremely low signal-to-noise ratio data.