The temporal magnetic disturbances which occur during magnetic repeat station surveys cause errors in the determination of the main field secular variation. However, their effects can be significantly reduced by optimizing the survey period and by modelling and subtracting the disturbance field. An analysis of data from five Canadian observatories, which form a meridional line from midlatitudes (Victoria) across the auroral zone to the polar cap (Resolute), shows that simple linear techniques for correcting the observations at a repeat station were generally unsuccessful. However, the subtraction of perturbations obtained by modelling data from 19 North American observatories using the new method of spherical cap harmonic analysis removed most of the temporal variations. After the removal of such fields, 80% of the errors at auroral zone locations were generally less than 10 nT. If repeat station surveys in high latitude regions are extended to several days, in order to increase the probability of observing a quiet or unsettled interval, and the disturbance field is modelled and removed from the observations, the errors in the secular variation are comparable to those obtained using observatory data.