Selections of events from seismicity catalogs on the basis of size are universally made on the basis of magnitude. Such a selection implicitly assumes that these magnitudes provide a temporally consistent measure of earthquake size or that systematic errors in magnitudes are constant. Several techniques can be used to test this assumption. Results of application of these techniques to local and teleseismic catalogs indicate that in many cases this assumption cannot be justified. Two different techniques for identifying changes in systematic errors in magnitude are described here and applied to recent seismicity data from Parkfield, California. One technique relies on comparisons of seismicity rates before and after a possible change in magnitudes. The other uses station magnitude corrections to redetermine magnitudes. Both of these techniques indicate that a systematic decrease in magnitudes by 0.18 to 0.19 units occurred at Parkfield during November 1984. This decrease was associated with the installation of six low-gain stations in the Parkfield region. It is clear that changes in stations distributions can have significant effects on magnitudes and that these changes must be accounted for in all seismicity studies which use magnitude as a selection criteria. The results of past studies which have not accounted for these changes must be reevaluated.