Concurrent with the explosion in the number of publications reporting biomarker discovery by profiling technologies, such as proteomics and pattern recognition, has been the increase in evidence highlighting the susceptibility of these approaches to analytical and experimental bias. The work presented here addresses these timely issues by delivering a detailed characterization of the effect of common sources of bias in clinical studies on serum and plasma profiles generated by a key technology in metabonomics, NMR spectroscopy. Specifically, differences in composition when blood samples were collected onto and in the absence of ice, over a series of serum-clot contact times, the stability of NMR-prepared samples over time and the effect on the metabolic profile of freeze-thawing were examined. While differences between individuals were far greater than variation from any other experimental factor, each of the conditions examined did cause slight alterations to the NMR profile that could produce a systematic bias. Variation due to clotting time caused changes in energy metabolites, which were delayed by ice with no other spectral effects. Room-temperature stability and hence NMR spectral repeatability were high (<1% intrasample variation). Higher molecular weight species such as lipoproteins were more susceptible to the variations present in the examined factors. These observations have implications for profiling study design, and hence, our results form a new and valuable resource for those attempting clinical metabolic profiling, for regulatory agencies involved in the licensing of clinical tests and in the generation of international reporting standards for metabonomics.