Statistical analysis was executed using The Unscrambler Version 7.01 (CAMO A/S, Norway). The approach of using multivariate techniques to correlate spectral data (in this instance near infrared spectra) with chemical and physical properties of a variety of biological materials is prolific in the literature. The general techniques are those of princi- pal component analysis (PCA), classical partial least squares with one response variable, y, (PLS-1) and pro- jection to latent structures (PLS-2) which models several response variables simultaneously and hence takes ad- vantage of any collinearity between the y-variables. De- tailed explanations can be found in Wold et al. (1987) and Martens and Næs (1991).