Osteoarthritis (OA) is a highly prevalent joint disease. Its slow progressive nature and the correlation between pathological changes and clinical symptoms mean that OA is often well advanced by the time of diagnosis. In the absence of any specific pharmacological treatments, there is a pressing need to develop robust biomarkers for OA. We have adopted a nuclear magnetic resonance (NMR)-based metabolomic strategy to identify molecular responses to surgically induced OA in an animal model. Sheep underwent one of three types of surgical procedure (sham (control), meniscal destabilization, MD or anterior cruciate ligament transaction, ACLT), and for every animal a serum sample was collected both pre- and postoperatively, thus, affording two types of "control" data for comparison. 1D 1H NMR spectra were acquired from each sample at 800 MHz and the digitized spectral data were analyzed using principal components analysis and partial least-squares regression discriminant analysis. Our approach, combined with the study design, allowed us to separate the metabolic responses to surgical intervention from those associated with OA. We were able to identify dimethyl sulfone (DMSO2) as being increased in MD after 4 weeks, while ACLT-induced OA exhibited increased 3-methylhistidine and decreased branched chain amino acids (BCAAs). The findings are discussed in the context of interpretation of metabolomic results in studies of human disease, and the selection of appropriate "control" data sets.