In this paper, we propose a powerful method of estimating the model parameters for time synchronization in wireless sensor networks (WSNs). Joint estimation of clock offset and clock skew has been proposed in the literature using the standard regression framework. Here, we claim that simple regression poorly estimates the parameters because of the inherent correlation among successive time readings between two sensors. We propose an alternative autoregressive model and use generalized least squares for estimating the relative offset and skew parameters. A computationally efficient Bayesian approach is also proposed for the parameter estimation considering correlated readings between two sensors. The effectiveness of the proposed approach compared with the earlier approach has been investigated through extensive simulation studies.