Sample matrices interfering with analyte determinations, termed matrix effects, are one of the factors limiting the more widespread use of environmental immunoassays. Previous attempts to reduce matrix effects have focused on particular assays in specific matrices rather than on general methods. Here we describe a novel method to eliminate one class of matrix effects in immunoassay, independent of the particular matrix or analyte. The method is demonstrated with a model system detecting estradiol in either a 10% methanol or a 5% dimethyl sulfoxide matrix. Fluorescently labeled antiestradiol antibody is introduced as the detecting antibody and excess unlabeled antiestradiol antibody is included as a reference antibody. The binding of the excess reference antibody to the sample analyte artificially creates a sample containing no free analyte to bind to the detecting antibody. This allows estimation of the fluorescent signal for "zero" analyte in the actual sample matrix. The solvents employed as model systems reduce the affinity of the detecting antibody and cause false positive results at low estradiol concentrations and false negative results at high concentrations. The proposed reference method, including addition of the reference antibody, resulted in a self-calibrating assay in which the matrix effects, both positive and negative, were completely eliminated.