A sensitive, robust Principal Component Analysis (PCA) method was developed to identify bacteria by their unique Surface Enhanced Raman Scattering (SERS) spectra.  SERS of bacteria has been previously demonstrated to produce species and strain specific spectra, including those recovered from blood.  These spectra are taken on a novel SERS substrate, consisting of Au or Ag 80-100nm spheres clustered on a glassy SiO₂ substrate.  PCA, a multivariate analysis tool, linearly projects a data set onto a set of orthogonal principal components (PCs), the basis, resulting from covariance matrix diagonalization.  The PCA of normalized SERS spectra was found to produce false negative classifications of bacterial strain and species, due to intensity variations resulting from baseline effects.  To emphasize SERS spectral peak location, a Heaviside function was applied to the negative second derivative of the spectra after noise removal via Fourier transform filtering.  The PCA of the resulting set didn't show intensity dependence and each spectrum was classified to its proper species/strain.  The distance from each centroid of all species clusters, in PC3 vs. PC2, to the nearest Voronoi class separator is at least three times the cluster standard deviation in our test data sets consisting of spectra from bacteria grown in broth and recovered from blood.  Following a normal distribution of data sets, the result implies at least 99.7% of all spectra acquired will be properly classified.  The high accuracy and 30-minute pathogen extraction to diagnosis time would make SERS an effective tool for clinical diagnostic settings and potential bio-threat identification.