Protein based biopharmaceuticals are becoming increasingly popular therapeutic agents. Recent changes to the legislation governing stem cell technologies will allow many further developments in this field. Characterisation of these therapeutic proteins poses numerous analytical challenges. In this work we address several of the key characterisation problems; detecting glycosylation, monitoring conformational changes, and identifying contamination, using vibrational spectroscopy. Raman and infrared spectroscopies are ideal techniques for the in situ monitoring of bioprocesses as they are non-destructive, inexpensive, rapid and quantitative. We unequivocally demonstrate that Raman spectroscopy is capable of detecting glycosylation in three independent systems; ribonuclease (a model system), transferrin (a recombinant biopharmaceutical product), and GFP (a synthetically glycosylated system). Raman data, coupled with multivariate analysis, have allowed the discrimination of a glycoprotein and the equivalent protein, deglycosylated forms of the glycoprotein, and also different glycoforms of a glycoprotein. Further to this, through the use of PLSR, we have achieved quantification of glycosylation in a mixture of protein and glycoprotein. We have shown that the vibrational modes which are discriminatory in the monitoring of glycosylation are relatively consistent over the three systems investigated and that these bands always include vibrations assigned to structural changes in the protein, and sugar vibrations that are arising from the glycan component. The sensitivity of Raman bands arising from vibrations of the protein backbone to changes in conformation is evident throughout the work presented in this thesis. We used these vibrations, specifically in the amide I region, to monitor chemically induced protein unfolding. By comparing these results to fluorescence spectroscopy and other regions of the Raman spectrum we have shown that this new method provides improved sensitivity to small structural changes. Finally, FT-IR spectroscopy, in tandem with supervised machine learning methods, has been applied to the detection of protein based contaminants in biopharmaceutical products. We present a high throughput vibrational spectroscopic method which, when combined with appropriate chemometric modelling, is able to reliably classify pure proteins and proteins 'spiked' with a protein contaminant, in some cases at contaminant concentrations as low as 0.25%.