Partial Least Squares modelling techniques have been used to calibrate Raman spectra in terms of the density of poly(ethylene terephthalate), with a standard prediction eff or of 0.0024 g cm-3. The calibration, which utilizes, primarily, the bands at approximately 1730, 1094, 997, and 860 cm-1, spans isotropic, uniaxial and biaxially oriented samples with a single model; this is superior to univariate calibrations which require separate models for isotropic and oriented samples. Hierarchical Cluster Analysis and Principal Components Analysis were also used for exploratory analysis of the spectra; these methods grouped samples in terms of their density and orientation characteristics, and also identified the bands which responded most significantly to morphology changes. It was shown that the 1615 cm-1 ring-stretching mode differentiated the samples on a basis other than simply by density or orientation alone; further work, however, is needed to understand this behaviour. The potential power of multivariate analysis techniques for correlating Raman spectra and polymer morphology was clearly demonstrated.