The Svalgaard-Mansurov (S-M) effect is used to derive the daily polarities of the interplanetary magnetic field (IMF) in the period from 22 May 1965 to 12 December 1998. The study analysis is based on measurements of the vertical component of the geomagnetic field carried out at Thule (near the geomagnetic North Pole) and at Vostok (near the geomagnetic South Pole). To avoid the uncertainties affecting most of the existing sets of inferred IMF polarities, we built up an algorithm which allows to infer them in a systematic way. We found that when IMF polarities are inferred from a single geomagnetic observatory, there are three phenomena that limit the accuracy of the results. They arise from the variability of the S-M effect in time. In fact, the signal: (1) has a polarity dependent amplitude (at Thule), i.e., it is slightly different in periods of opposite polarities; (2) is seasonal-dependent, being stronger during the local summer than during winter; (3) depends on the geomagnetic activity level. The last effect favors the detection of positive polarities at north (Thule) and of negative polarities at south (Vostok), during periods of enhanced geomagnetic activity. The success rates of the inferring method turn out to be rather high (93.9% for Thule and 86.6% for Vostok), but the time coverage is poor (73.3% for Thule, 49.7 % for Vostok). If one then combines simultaneous data from the two stations, results improve substantially: in fact, an overall success rate of 95.6% is obtained, along with a coverage of 93.0% of the time. The inferring method was then applied to the presatellite era obtaining the IMF polarities for the period 1958--1965 when magnetograms from Thule and Vostok were simultaneously available.