Blood serum and plasma are the most frequently used biofluids in metabolomics. The primary aim of this study was to ascertain the feasibility of the commonly used gas chromatography-mass spectrometry (GC-MS)-based methoximation followed by silylation with the BSTFA + 1% TMCS (MeOx-TMS) derivatization method, and to reveal the metabolic differences between serum and plasma. Individual variations were evaluated by different groups of serum and plasma samples collected from healthy volunteers in 2011 and 2013. The experimental results indicated that differences of metabolic levels among individuals were much higher than the variations of the experimental repeatability and precision. In addition, a discriminant model between serum and plasma was established using partial least squares-discriminant analysis (PLS-DA). Six characteristic metabolites, phosphate, serine, 2,3,4-trihydroxybutyrate, citric acid, glucose and arachidonic acid, were screened out and considered to have the most important contribution to the discrimination. Results of this work will provide some valuable suggestions to researchers on the selection of suitable biofluids in metabolomic research.