Abstract: In modern cities, taxis origin-destination data is a kind of very useful big traffic data, which contains rich spatio-temporal information. This paper designs a visual analysis system for mining spatio-temporal patterns of big taxis origin-destination data. The system helps users determine the regions of interest for further investigation by the global overview of origin-destination distributions; uses the circular pixel graph to visually encode the spatio-temporal patterns of the origin-destination data of the analyzed regions chosen by lasso or rectangle region selection tools; and provides multi-dimensional analysis of the latent spatio-temporal patterns of taxis origin-destination data through interactions between multiple coordinated views of visualizations. The designed system obtains some interesting findings and receives good user feedbacks when applied to real taxis GPS data in Hangzhou. The system is not only useful for vehicle administration departments to dispatch taxi cabs according to temporal demands but also helpful for taxi drivers to earn more money.