A high number of spectra of samples from a metabonomic study render an automated data analysis necessary. In addtion, NMR spectra of a set of samples from a metabonomics study are highly complex. Systematic differences between samples are often hidden behind biological noise and behind shifting peaks. Therefore robust methods for data mining are needed. Although first approaches for an automated fitting of spectra have been proposed recently, practically all publications of NMR metabonomics in literature use data decompositon methods such as principal component analysis (PCA), partial least squares (PLS) or othogonal partial least squares (O-PLS). These methods look for (systematic) variances between samples. In contrast to the most popular method PCA, O-PLS and PLS use information about samples (such as dosegroup or dose). Therfore, these methods allow often a better separation of samples and a clearer identification of significant variables, but are biased in contrast to PCA, which is explained in more detail in this section.