The maximum a posteriori classifier as applied to spectral parameters has been found to successfully classify radar clutter. Error rates of less than 10% were generally found using recorded radar data. The feature sets were computed using the multisegment Burg formula and the lattice implementation of the prediction error filter, followed by the application of spectrum shifting. This classifier is also found to successfully assign undefined clutter types to their most closely related classes. The various clutter types are reviewed together with the problems of applying spectral analysis to the radar returns. Next, the necessary feature set is derived from the set of reflection coefficients, followed by the application of Bayes' decision theory for their classification.