Extraction of seismic wavelets is important for impedance inversion and forward modeling. We propose a method for estimating spatial variant seismic wavelets. The method uses third- and fourth-order cumulants of seismic data. It combines third- and fourth-order moments of seismic wavelets with those obtained by a deterministic method at well locations. We design a multidimensional filter at a well by matching the high-order cumulants of seismic data to the high-order moments of seismic wavelets. Applying the filter to the high-order culumants of other seismic traces away from the well yields the corresponding wavelets. The new method for seismic wavelet extraction is thus constrained by well-log data. Synthetic tests show that this method can produce accurate spatially varying wavelets. A real data test of the method is also successful.