Travel time is a critical measure for road network traffic conditions, and travel time estimation provides available information for travellers and traffic management. This paper proposes an improved method based on Markov Chains to estimate route travel time by considering spatio-temporal correlation from related links. The method mainly contains three parts. Firstly, in the light of traffic flow data collected from microwave detectors, Gaussian mixture model (GMM) is applied to cluster travel time data under two consecutive links, and thus capture the underlying traffic states. The transition probability matrix is constructed to estimate variations of traffic states over time. Then, link travel time distributions can be estimated from historical observations. Accordingly, we can estimate route travel time distribution by aggregating weighted link travel time distribution based on convolution theory. Finally, a case study including three experiments are used to test the accuracy of travel time estimation, we also compare the estimation performance of proposed model with several traditional methods, and the results indicate that the proposed model is effective and superior to traditional modes based on two indicators: Kullback–Leibler (KL) divergence and Mean Absolute Error (MAE).