The comparison of analytical results calls for validation of the assays used, i.e. for the documentation of accuracy (trueness and precision), linearity and specificity. In addition, there is a growing demand for evaluation and documentation of traceability and uncertainty of analytical results. However, models for establishing the traceability and uncertainty of immunoassay results are lacking. Sandwich enzyme-linked immunosorbent assays (ELISAs) were developed for determination of the human cytokines interleukin-4 (IL-4), interleukin-5 (IL-5), interferon-y (IFN-gamma) and tumor necrosis factor-alpha (TNF-alpha). The accuracy of each of the assays was evaluated in the ranges of 1-15 microg/l (IL-4), 0.001-1 microg/l (IL-5), 0.5-2.5 microg/l (IFN-T) and 0.14-2.2 microg/l (TNF-alpha). Other evaluated performance characteristics were the limit of detection (LOD), immunological specificity, and robustness. Traceability was ensured by the use of World Health Organization International Standards (WHO IS). An uncertainty budget, which combined the contribution from all known uncertainty components, was established for each cytokine ELISA. The between-run relative analytical standard deviation (RSDA) of the assessed ELISAs was found to be in the range of 11-18%, except for IL-5 where RSDA increased at decreasing concentrations. The LOD was 0.12 microg/l, 0.0077 microg/l, 0.0069 microg/l and 0.0063 microg/l for IL-4, IL-5, IFN-gamma and TNF-alpha, respectively. Traceability to the WHO IS was established for each of the cytokines. The combined relative standard uncertainty (U(result)/C(result)) was 28%, 22-62%, 28% and 24% for the IL-4, IL-5, IFN-gamma and TNF-alpha results, respectively. The major contributions to uncertainty came from the relative analytical standard deviation and from the uncertainty of the mass concentration of the WHO IS. The uncertainty of the WHO IS was not stated in the accompanying certificate and was evaluated by other means. The largest sources of uncertainty were located outside our laboratory. This means that the possibilities to improve the reliability of the results produced by the ELISAs are very limited. The task of evaluating measurement uncertainty would be much easier if producers of international reference standards reported the uncertainty of the value of standards. The model for evaluating uncertainty presented in this paper is applicable to other types of assays and to most analytical methods.