Several remarks made by Stark [1996> are in basic agreement with those of Varotsos et al. [1996a> (e.g., ''If we choose to issue a prediction only when the expected magnitude exceeds 5.0, then, if our prediction algorithm works, we would expect to fail to predict some events with magnitude 5.0 and smaller (and even some larger events)'', ''It is generally accepted that ''raw'' seismicity series are not Poisson distributed...'', etc.). However, in this Reply we clarify a few misunderstandings that led Stark [1996> to state that Varotsos et al. [1996a> made some erroneous suggestions. We emphasize that the tolerance limits in the big majority of the VAN predictions were not calibrated a posteriori, because these limits were published one year before the period 1987--1989 under discussion. Only in two, out of 25, successful correlations the Δt-value was extended, a posteriori; we emphasize, however, that these two predictions were recognized well in advance as belonging to a new case which was then labelled as SES electrical activity (sequence of SESs) that differs from the case of single (isolated) SES. We do agree with the Stark's [1996> suggestions according to which one ''avoids the necessity of specifying a probability distribution for earthquake variables, a task that is both controversial and problematic.'' ¿ American Geophysical Union 1996