A novel near infrared spectroscopic technique was used to characterize the joints in arthritis with comparison against normal joints. A beam of near infrared light was passed to joints through a fibre optic cable. Scattered light was collected by the same fibre bundle and a spectrum of the joint computed. Multivariate pattern recognition techniques identified regions of the spectrum which allowed discrimination between healthy and affected joints. Linear discriminant analysis resulted in correct classification of 74% of the joints. The high degree of similarity between mean spectra representing the early, late and control groups along with the significant between—subject variability in the data make diagnosis based on visual assessment of the spectra impossible. Linear discriminant analysis was therefore applied to spectra to determine if spectra could be classified by statistical methods as arising from early or late RA. Application of LDA resulted in correct classification of 74% of the joints. Interestingly, the spectral regions in which diagnostic differences were found by the multivariate analysis contain absorption bands related to tissue oxygenation status (oxy and deoxyhaemoglobin) and oxygen utilisation (cytochrome aa3), suggesting that ischaemic changes within the joint are being detected.