A comparative study has been accomplished between diagnostic mediums: human oral tissue and saliva for oral cancer detection on three groups: oral squamous cell carcinoma (OSCC), dysplastic, and control (normal) by using Stokes shift (SS) spectroscopy (SSS) at ▵λ of 120 nm, which is the Stokes shift of nicotinamide adenine dinucleotide (NADH). SS spectra obtained from tissue and saliva consist of major bands of collagen, tryptophan, NADH and minor bands of flavin adenine dinucleotide (FAD) and porphyrin. Principal component analysis (PCA) has been performed on the data sets of SS spectra for discrimination by dimension reduction. Linear discriminant analysis (LDA) has been applied on the PC scores to compute linear discriminant (LD) scores. Kernel probability density functions of LD scores are plotted to show how LD scores of each group are separated from one another. LD scores for oral tissue differentiates OSCC to normal, dysplasia to normal, and OSCC to dysplasia with sensitivities 100 %, 85 %, 94 % and specificities 88 %, 88 %, 89 % with the accuracy of 95 %, 87 % and 92 %. On the other hand for saliva, it differentiates respective groups with sensitivities 91 %, 82 %, 91 % and specificities 100 %, 88 %, 82 % with the accuracy of 95 %, 85 % and 87 %. Obtained results with human saliva are as prominent as oral tissue and we conclude that it may be used as a substitute diagnostic medium. In addition SS spectroscopy instead of fluorescence spectroscopy at 120 nm shift appears to be an important tool for in vivo detection of early oral cancer.