In this paper, we examine how variations in normal tissue can influence disease classification of Raman spectra. Raman spectra from normal areas may be affected by previous disease or proximity to areas of dysplasia. Spectra were acquired in vivo from 172 patients and classified into five tissue categories: true normal (no history of disease), previous disease normal (history of disease, current normal diagnosis), adjacent normal (disease on cervix, spectra acquired from visually normal area), low grade, and high grade. Taking into account the various "normal" states of the tissue before statistical analysis led to a disease classification accuracy of 97%. These results indicate that abnormal changes significantly affect Raman spectra, even when areas are histopathologically normal. The sensitivity of Raman spectroscopy to subtle biochemical differences must be considered in order to successfully implement it in a clinical setting for diagnosing cervical dysplasia and cancer.