Metabolomics lies at the fulcrum of the system biology 'omics'. Metabolic profiling offers researchers new insight into genetic and environmental interactions, responses to pathophysi- ological stimuli and novel biomarker discovery. Metabolomics lacks the simplicity of a single data capturing technique; instead, increasingly sophisticated multivariate statistical techniques are required to tease out useful metabolic features from various complex datasets. In this work, two major metabolomics methods are examined: Nuclear Magnetic Resonance (NMR) Spec- troscopy and Liquid Chromatography-Mass Spectrometry (LC-MS). MetAssimulo, an 1H-NMR metabolic-profile simulator, was developed in part by this author and is described in the Chap- ter 2. Peak positional variation is a phenomenon occurring in NMR spectra that complicates metabolomic analysis so Chapter 3 focuses on modelling the effect of pH on peak position. Analysis of LC-MS data is somewhat more complex given its 2-D structure, so I review existing pre-processing and feature detection techniques in Chapter 4 and then attempt to tackle the issue from a Bayesian viewpoint. A Bayesian Partition Model is developed to distinguish chro- matographic peaks representing useful features from chemical and instrumental interference and noise. Another of the LC-MS pre-processing problems, data binning, is also explored as part of H-MS: a pre-processing algorithm incorporating wavelet smoothing and novel Gaussian and Exponentially Modified Gaussian peak detection. The performance of H-MS is compared alongside two existing pre-processing packages: apLC-MS and XCMS.