Raman spectroscopy can probe the chemical structure of a material providing an optical ’fingerprint’ unique to the sample. Such is the capacity of Raman spectroscopy to identify differentmaterials, it be to classify biological cells and tissue and can provide an ’optical biopsy’ for various types of disease. A key component in Raman diagnostics is the use of multivariate statistical algorithms that can be trained using datasets of known samples to classify the groups based on the subtle differences between them. Despite the great progress in this field in recent decades, Raman spectroscopy has never been adopted clinically. The key reason for this is the poor resproducibility of Raman spectroscopy across instruments; in other words the same material can produce different spectra when recorded using different spectrometers. These differences can include small movement of the Raman peaks along the wavenumber axis (wavenumber miscalibration) or modulation in the amplitude of the peaks (intensity calibration). Such changes can render a multivariate classifier trained on one instrument to be completely useless in identifying samples recorded from another instrument. The overall goal of this thesis is to develop new methods for wavenumber and intensity calibration that can help Raman spectroscopy penetrate into the clinic.