Enzyme immunosorbent assays (ELISA) for human tumor necrosis factor-alpha (TNF-alpha) are available from many companies. Although this cytokine is meanwhile well established in scientific work its detection in human blood still leads often to conflicting results. We studied the analytical performance of three different commercial ELISA (Amersham, Immunotech and Medgenix Diagnostics) in plasma samples of 30 human immunodeficiency virus (HIV) infected individuals. All kits showed significantly different mean concentrations of TNF-alpha in the study population (p < 0.05). Kit A: 30.9 pg/ml +/- 46.6 vs. kit B: 46.9 pg/ml +/- 24.8 vs kit C: 11.8 pg/ml +/- 11.6. Results of assays A and B (r = 0.34) as well as A and C (r = 0.31) were not significantly correlated. A good statistical relation between values was only found for assay B and assay C (r = 0.80. p < 0.001). Three plasma samples were spiked with TNF-alpha standard of the corresponding kit to a final concentration of 200 pg/ml. The detection results of these samples were compared with the variation conceded by the manufacturer. Assay A detected two samples above the corresponding coefficient of intra-assay variation (recovery: 81.8% and 89.2%). All three values obtained with assay B were markedly out of its variation range (recovery: 88.7%, 84.6% and 89.0%) while assay C showed again two results above the intra-assay variation (recovery: 92.1% and 89.6%). In our observations we found relevant differences between commercial ELISA kits of different manufacturers, which made interpretation of TNF-alpha in human plasma samples difficult. Additionally, it should be taken into consideration that plasma specimen may contain cytokine-binding proteins or autoantibodies which are capable of blocking epitopes of TNF-alpha. This may lead to the loss of the necessary immunoreactivity which prevents detection antibodies from recognizing these molecules.