The estimation of travel time is one of the key factors of traffic guidance and transportation management. However, it is impossible to collect the time of passing all the roads during the data collection phase. Due to the sparseness of data, the prediction accuracy of existing methods is usually not satisfactory. In this paper, we consider introducing the concept of tensor to make full use of spatial-temporal traffic patterns. A model based on tensor decomposition for estimating the travel time from the citywide perspective is proposed. This model is comprised of three major components: travel time tensor construction, travel time tensor factorization and unconstrained optimization. Different dates, different time periods, and different road segments are modeled as three dimensions of tensors. In this paper, an example analysis of the monitoring data collected by Ruian City in China is carried out. In addition, this paper verifies the conditions in which the data is missing at different degrees, the accuracy of the model. The results show that the model can achieve great prediction accuracy.

CICTP