A systematic error is known to occur in estimating surface-wave group-velocity dispersion when the multiple filter analysis technique is applied. The error comes from the fall-off of the amplitude spectra at long periods. We propose a method for correcting this error. It consists of using a centroid frequency of the filtered spectrum instead of the central frequency of the gaussian window. The method is especially useful when the group-velocity curve is obtained from stacking of individual frequency-time diagrams. We apply this technique to two data sets. The first one consists of nine seismograms of coastal, subduction-zone earthquakes recorded by a broadband station located in Mexico City. This data set has been previously used to estimate an average crustal structure of southern Mexico. The second data set consists of broadband seismograms recorded in India and has been used to determine an average structure of the Indian peninsular region. Our results show that, in the first case, the systematic error is negligible. This is due to the relatively low decay rate of the spectral amplitudes at long periods. However, in the case of the Indian data, the systematic error of the multiple filter analysis cannot be neglected since it changes significantly the measured dispersion curves and leads to errors in the estimated crustal structure. Our tests show that the proposed method successfully corrects a major part of the systematic error.