Magnetotelluric (MT) data acquired during September-October, 1994, in northern Canada, were strongly influenced by non-uniform source field contributions from the auroral electrojet, and especially by intense auroral episodes. The largest effect on the estimate of the magnetotelluric impedance tensor elements was during intervals of highest magnetic activity, which primarily correlated with high auroral activity and was observed during local nighttime. In comparison, during the day the effect on the normal magnetotelluric impedance tensor response was usually, but not always, small. A robust controlled-leverage processing algorithm was applied to these data in an attempt to extract the stable uniform field estimates of the impedance. The differences between nonrobust and robust processing of the entire data set is compared to that obtained after dividing the time series into daytime and nighttime segments. The nonrobust estimate using all data is controlled by the nocturnal data, which are, in turn, dominated by non-uniform source effects. However, nonrobust processing of only the daytime data fails to recover a useful result. There is little difference between the robust response for the entire and daytime data provided that the fraction of auroral activity is not large, i.e. in excess of half of the available data series. In addition, examination of the time-dependence of the response functions shows that the strongest bias is observed during the initial quarter of an auroral event.