We report the development of a probability-based multi-class diagnostic algorithm to simultaneously distinguish high-grade dysplasia from low-grade dysplasia, squamous metaplasia as well as normal human cervical tissues using near-infrared Raman spectra acquired in-vivo from the cervix of patients at the Vanderbilt University Medical Center. Extraction of diagnostic features from the Raman spectra uses the recently formulated theory of nonlinear Maximum Representation and Discrimination Feature (MRDF), and classification into respective tissue categories is based on the theory of Sparse Multinomial Logistic Regression (SMLR), a recent Bayesian machine-learning framework of statistical pattern recognition. The algorithm based on MRDF and SMLR was found to provide very good diagnostic performance with a predictive accuracy of ∼90% based on leave-one-out cross validation in classifying the tissue Raman spectra into the four different classes, using histology as the "gold standard". The inherently multi-class nature of the algorithm facilitates a rapid and simultaneous classification of tissue spectra into various tissue categories without the need to train and heuristically combine multiple binary classifiers. Further, the probabilistic framework of the algorithm makes it possible to predict the posterior probability of class membership in discriminating the different tissue types.