The problem of estimating the timing of ultra-wide band signal is considered in the letter. We develop a maximum likelihood timing estimation algorithm for binary PAM DS-UWB systems. The derivation of the proposed algorithm is based on the known training sequence and AWGN channel. The Cramer-Rao Bound (CRB) for the ML timing estimator is presented as a performance benchmark. It is found via numerical results that the ML timing estimator on AWGN channels achieves the CRB when the values of Eb/N0 for the observation bits Nb=50 are sufficiently high. Finally, the performance of the proposed ML estimator is evaluated on actual channels with intersymbol interference such as an IEEE UWB indoor multipath channel model.