The science of metabolomics is a emerging field that requires intensive signal processing and multivariate data analysis for interpretation of experimental results. While any technique that quantifies metabolites can be used for metabolomics, the work presented herein is focused on the data processing and analysis of nuclear magnetic resonance (NMR) spectroscopic data sets. Typically, these large scale data sets are analyzed as follows: (1) standard post-instrumental processing of spectroscopic data; (2) quantification of spectral features; (3) normalization; (4) scaling; and (5) multivariate statistical modeling of data. Each one of these computationally intensive steps can be further subdivided into individual algorithms, where the researcher must select from several competing algorithms. Selecting the best technique for a given task is dependent on the goals of analysis, and therefore, any analysis platform must be robust and flexible. In addition, the parallel nature of processing steps is seldom realized. This work presents a service oriented scientific workflow approach to NMR-based metabolomics data analysis. We demonstrate the effectiveness of this approach by implementing several common spectral processing techniques in the cloud using a parallel map-reduce framework, Hadoop. Due to its mapreduce architecture and its fault-tolerant file system, Hadoop is ideal for analyzing large spectroscopic data sets. The adoption of a scientific workflow via Taverna allows the flexibility to select the most appropriate data analysis technique, regardless of their implementation details. The advantages of this approach are demonstrated on a NMR-based metabolomics dataset from a rat model of tissue toxicity.