Abstract. The aim of this paper is to lead a practical, rational and rigorous approach concerning what can be done, based on the knowledge of magnetic series, in the field of prediction of the extreme geomagnetic events. We compare the magnetic vector differential at different locations computed with different resolutions, from an entire day to minutes. We study the classical correlations and the simplest possible prediction scheme to conclude a high level of predictability of the magnetic vector variation. The results obtained are far from a random guessing: the error diagrams are either comparable with earthquake prediction studies or outperform them when the minute sampling is used in accounting for hourly magnetic vector variation. We demonstrate how the magnetic extreme events can be predicted from the hourly value of the magnetic variation with a lead time of several hours. We compute the 2-D empirical distribution of consecutive values of the magnetic vector variation for the estimation of conditional probabilities of different types. The achieved results encourage further development of the approach to prediction of the extreme geomagnetic events.