Abstract: With the increasing progress on GPS, RFID and wireless technologies, moving objects are becoming increasingly attractive to data mining community. Trajectory outlier detection has many practical applications, so it has been a popular data mining task. Existing trajectory outlier detection algorithms are sensitive to some parameters and need to manually adjust the parameters, which lead to some drawbacks of unstable and less robustness. Furthermore, most of existing algorithms detect outlying trajectories only according to self-defined dimension and use less information of known trajectories to assist in detecting more potential outlying trajectories. A new trajectory outlier detection algorithm based on semi-supervised technology is proposed in this paper. It makes full use of the information of this little known trajectory, which include the distribution of the known data and the user’s understanding of abnormal data. We use them to calculate sensitive key parameters by semi-supervised technology, and overcome the instability of existing algorithms which comes from sensitive key parameters. The algorithm detects valuable outlying trajectories with new dimension at global and local views. Experimental results show that the proposed algorithm can discover more valuable outlying trajectories with less parameter adjustment.