The geoid error, computed from EGM08, is unrealistically large due to the continuation of the spherical harmonic coefficient errors down to the surface of the reference ellipsoid. In this study, we try to calibrate such an error by the differences between the EGM08 and GNSS/levelling geoids over Fennoscandia. We use the variance component estimation procedure through combined adjustments of the geoid and GNSS/levelling heights using corrector surfaces of 4-, 5- and 7-parameter. We also develop a simple iterative method to calibrate the geoid error from the a posteriori variance factor and the errors of GNSS/levelling geoid. Our numerical investigations show that performing the separate adjustment and variance component estimation for each country with a two-component stochastic model is more successful than performing it in the whole area with a five-component model. The number of GNSS/levelling data over Sweden and Norway are much larger than those in Denmark and Finland. This causes that the corrector surfaces are fitted better in these countries and consequently the estimated errors for the geoid become larger than what they should be in the others. Based on a 7-parameter corrector surface model, the average error of the EGM08 geoid becomes 12, 17, 51 and 34 mm, in Sweden, Denmark, Norway and Finland, respectively. If the two-component stochastic model is used in a combined adjustment over Fennoscandia this average error will be 48 mm.