The analysis of two 91Zr neutron transmission experiments performed some years ago at the Oak Ridge National Laboratory and Geel electron Linacs showed systematic discrepancy in the resonance neutron width estimation. The two sets of data are comprehensively reanalyzed by a shape least-squares method that builds up a single system of normal equations. An accurate resolution function description was obtained by Monte Carlo calculation of the moderation and detection contributions. The two experiments are found to be in complete agreement and a single set of resonance neutron widths is obtained.