It is very difficult to estimate the relative arrival delay of the eigenrays for an unknown source in shallow water. The effects of a source position changing in the neighborhood and sound speed profile perturbation on arrival time of eigenvays are similar. In this paper, we present a robust localization method based on the auto-correlation function of wide-band signal of single hydrophone. By designing neighboring location constraints, a weighting function is constructed to change the peak cross-interference of the auto-correlation function to useful information that is conducible to the improving of targeting performance. In this method there is no need to estimate the relative arrival delay of the eigenray. Computer simulation shows that the robust method can achieve better localization performance, and even has tolerences of environment mismatch and searching grid mismatch. The performance of the robust method is validated through the broad-band data collected on a vertical line array during the Shallow Water 2006 experiments.