A design of experiments method based on the criterion of D-optimality is presented. It is applied to the estimation procedure of the parameters of a periodate calibration curve described by a model equation of the form y =ax 2 bx + cx + 1) and giving the maximum possible precision using three experimental measurements. The use of four or five experimental measurements for producing the calibration curve was not found more advantageous than that of three. Simulated data were generated, for the three optimal points, and a large number of parameter estimates were obtained under constant and/or relative errors (in y) showing a robust behaviour. The %-bias and the %-excess variance based on the simulated data were found to be non-significant.