Seismic reflection data are usually recorded with sample intervals of 2 or 4 msec. It is, however, apparent from sonic and density logs that seismic reflections occur at much shorter intervals. Thus, even the narrowest signal possible with a Nyquist frequency of 250 or 125 Hz averages over many reflections. Under these circumstances it appears to be more advantageous to design deconvolution filters which place more emphasis on the reduction of side lobes than on the sharpening of the deconvolved pulse. Given the signal-to-noise (S N) ratio of a seismic trace with known wavelet, a frequency domain filter can be computed which leads to a higher output S N ratio than its standard counterpart at the expense of a slightly wider peak. The wider peak means that the deconvolved signal averages over more reflections but side lobes of earlier and later reflections and filtered noise combined are lower than on a standard deconvolved trace. A good agreement between the wavelet used for deconvolution and the actual wavelet appears to be important for successful deconvolution.