The performance of a sandwich-ELISA for TNF measurement in plasma and serum was studied. The ELISA was first statistically analyzed. Interassay coefficient of variance and the intraassay coefficient of variance for the concentration range between 0.5 and 5 ng/ml was less than 10%. The sensitivity of the sandwich-ELISA for TNF in culture medium was 10 pg/ml. The ELISA was shown to be specific for biologically active TNF, since a good correlation between the ELISA and the WEHI bioassay was observed when partially inactive, denatured TNF was measured. The effect of various anticoagulation systems on the reliability of human TNF measurement has been evaluated. The oxalate/NaF and EDTA systems were both appropriate, as appeared from the observed blockade of the production of TNF in the tube, either in the cell-glycolysis-blocked or in the calcium-depleted situation, respectively. An eventual decrease in the recovery of rTNF after collection of blood was prevented in the oxalate/NaF tubes. Recovery of TNF in the ELISA was diminished in the presence of plasma or serum. Techniques to enhance the efficiency of the measurement of TNF in plasma by ELISA were compared. The data indicate that in the presence of 1.1 M NaCl, the TNF masking effect of normal plasma was largely abrogated. The presence and role of inhibiting plasma components in plasma of healthy and diseased individuals are discussed.