The evaluation of seismic risk at locations where sensitive man-made structures are planned depends critically on a correct estimate of the maximum expectable earthquake magnitude, Mmax, in that region. By assuming that the longest fault (or fault unit) with length Lmax could break in a single earthquake, one estimates Mmax from Lmax on the basis of a magnitude versus source-length relation, which is derived empirically. The maximum expectable ground accelerations are then estimated from Mmax. I propose that a more accurate estimate of Mmax can be obtained by determining the maximum expected rupture area, Amax, and using the magnitude-area relation M = log A + 4.15 (valid for M > 5.6). Amax can be obtained from the product of Lmax times the expected fault width. The latter can probably be estimated more accurately than Lmax on the basis of tectonic analysis and microearthquakes studies. The Mmax estimates derived from rupture area give more accurate results than the estimates based on rupture length alone, because narrow faults produce less powerful earthquakes than do wide faults of the same length.