It has been considered that the p value in the modified Omori formula does not depend on the cut-off magnitude <Utsu et al., 1995>, and large and small aftershocks decay similarly <Reasenberg and Jones, 1989a, 1989b>. Having investigated temporal decrease of aftershock activities of the 1995 Kobe and the 2000 Tottori earthquakes in detail, we found, however, that large aftershocks do not have a stable decay pattern, the p value in early periods is smaller than that obtained by the data in long periods which contain the later stage of the sequence. To say the point in another way or more clearly, number of large aftershocks in later periods is significantly short compared to that expected from the model that is obtained by fitting the decay in early time periods to the modified Omori formula. Preliminary analyses for several other aftershock sequences suggest that this is a common feature observed in all aftershock activities.