A radioimmunoassay (RIA) for insulin was validated for reliable measurement of the human insulin analogue, insulin aspart, by correction of non-linear measurements. Specificity was equivalent for several species of insulin, except insulin aspart. A non-linear hyperbolic model fitted insulin aspart with a correction formula for non-linearity of: z = 1,503y/ (1,398 - y), where y denotes measured concentration and z denotes true concentration. Matrix-effects were insignificant for human, porcine, and canine heparin-plasma and for human and porcine serum. The coefficient of variation was below 15% for 80-800 pmol/L human and porcine insulin and for 80-600 pmol/L insulin aspart. The limit of detection for insulin aspart was 11.5 pmol/L with a lower limit of quantification of 17.5 pmol/ L. Dilution of serum with Pharmacia dilution media introduced no significant error. In conclusion, this paper demonstrates that a non-parallel radioimmunoassay can be used to estimate accurate concentrations of insulin aspart.