Attention is drawn to a bias occurring in the course of calibration. Since the usual assumption of normality leads to an estimate without expectation. here a class of distributions of the errors is introduced which does not cause such a deficiency. Even in this class. a correction of the regression line defined with the help of standards might be necessary before estimating the abscissa value belonging to an unknown object. The estimate introduced is compared lo the classical and inverse estimates. A mode of eliminating the error by experimental design is also discussed. The calibration procedure was simulated by computer.