Serum was drawn from each individual at the time of enrolment. The serum samples were kept frozen until measured. The samples were measured at the University of Nijmegen. The enzyme linked immunosorbent assay for the measurement has been published earlier . Briefly, microtiter plates (Nunc, Roskilde, Denmark) were coated overnight at 4 °C with rabbit antihuman IgD (200 mul, Dako, Copenhagen, Denmark; 1 : 5000 dilution in bicarbonate buffer, pH 9.5). After automatic washing of the plates with phosphate buffered saline (PBS) containing 0.01% Tween 20 (washing buffer), serum samples or standard serum dilutions (100 mul) diluted in washing buffer were added. These were also incubated overnight at 4 °C. After washing, the plates were incubated for 90 min at 37 °C with mouse monoclonal antihuman IgD recognizing the Fc-part of the IgD molecule (Dako; 100 mul 1 : 200 in washing buffer). After washing, Horse-Radish-Peroxidase labelled rabbit antimouse immunoglobulins were added (Dako; 100 mul 1 : 4000 in washing buffer) and incubated for 90 min at 37 °C. The plates were washed again and incubated with substrate containing orthophenylenediamine (Sigma, St Louis, MO, USA; 2 mg/ml) and H2O2. After stopping the reaction with 2 m H2SO4 absorbance was read at 492 nm, using a Titertek Multiskan ELISA reader (Eflab, Oy, Helsinki, Finland). As the standard serum, Behring OTRD 02/03 (Marburg, Germany) calibrated against the British Research Standard 67/37, was used. This standard serum contains IgD-Fc fragments. In this ELISA, only Fc-regions are measured and standardization is also performed on the Fc-regions. Presumably therefore, no influence of the notorious splitting of IgD molecules has been found in this assay. Even after storage for 24 h at room temperature still the same values were measured. The lower limit of detection of this ELISA was 1 IU/ml (1.4 mg/l). Special effort was made to read values as low as 0.5 IU/ml. Statistical methods. The nonlinear regression model a(1-e−b age)/(age + c) was adapted to the log of the IgD data against age by the SPSS-statistical package. The parameters a and b were evaluated for a few values of c. The value c = 20 gave the best fit with the model. The corresponding estimates of a and b with their 95% CI were a = 138.7 (122.4–155.0), b = 0.17 (0.13–0.21), c = 20. A change of c by, e.g. 2 units (± 2 units) gives estimates for a and b which are well within the respective confidence intervals given. The variation in the log of IgD explained by this model is 53% of the total variation. According to this model the predicted means of the IgD concentration, at each age, and its 95% percentile was calculated. IgD values below the detection limit, 0.5 IU/ml, of our method were given the value 0.25 IU/ml.