Oral cancers su®er from poor disease-free survival rates due to delayed diagnosis. Noninvasive, rapid, objective approaches as adjuncts to visual inspection can help in better management of oral cancers. Raman spectroscopy (RS) has shown potential in identi¯cation of oral premalignant and malignant conditions and also in the detection of early cancer changes like cancer-¯eld-e®ects (CFE) at buccal mucosa subsite. Anatomic di®erences between di®erent oral subsites have also been reported using RS. In this study, anatomical di®erences between subsites and their possible in°uence on healthy vs pathological classi¯cation were evaluated on 85 oral cancer and 72 healthy subjects. Spectra were acquired from buccal mucosa, lip and tongue in healthy, contralateral (internal healthy control), premalignant and cancer conditions using¯ber-optic Raman spectrometer. Mean spectra indicate predominance of lipids in healthy buccal mucosa, contribution of both lipids and proteins in lip while major dominance of protein in tongue spectra. From healthy to tumor, changes in protein secondary-structure, DNA and heme-related features were observed. Principal component linear discriminant analysis (PC-LDA) followed by leave-one-out-crossvalidation (LOOCV) was used for data analysis. Findings indicate buccal mucosa and tongue are distinct entities, while lip misclassi¯es with both these subsites. Additionally, the diagnostic algorithm for individual subsites gave improved classi¯cation e±ciencies with respect to the pooled subsites model. However, as the pooled subsites model yielded 98% speci¯city and 100% sensitivity, this model may be more useful for preliminary screening applications. Large-scale validation studies are a pre-requisite before envisaging future clinical applications.