A comparison is made between secular variation values computed at the positions of 168 magnetic observatories from two spherical harmonic models based almost entirely on the 7-month spread of Magsat data and observed secular variation values derived from the observatory annual mean values. The two models differ from each other (rms difference in Z=27.8 nT/yr) and from the observed values (rms differences in Z of 22.1 and 30.6 nT/yr). It appears that a time span of more than 7 months is needed to produce accurate models of the secular variation from satellite data.