Abstract Amplitude variation with offset (AVO) analysis is sensitive to errors in the moveout correction. To reduce moveout errors, AVO-effects are estimated using time windows. For each time window, block-NMO is used to avoid NMO-stretch, and correction for residual NMO is performed by an iterative crosscorrelation technique. The data in a moveout-corrected time window are modeled as a constant pulse multiplied by an amplitude function that is approximated by a polynomial in the offset coordinate. Conversion of data from offset to angle, or slowness, is therefore avoided. The seismic pulse and the polynomial coefficients are found by the least-squares method. This is a separable least-squares problem and the polynomial coefficients and the pulse can be estimated separately. The reduced nonlinear optimization problem for the coefficients is expressed as a Rayleigh quotient, providing the solution in one step. Once the coefficients have been found, the estimate of the pulse is computed by an explicit formula. This gives an efficient computational scheme. The method is demonstrated on a shallow seismic anomaly in the Barents Sea, offshore northern Norway.