Abstract: Due to the limitation of the present techniques and facilities for data collection and various interferences, the data obtained are often distorted and noised, directly influencing the result of subsequent data analysis. The conventional approaches to outlier removal either assume that the data follow a certain known distribution or deal with the data that are from a single distribution, resulting in a reduced credibility of the data processed. This paper proposes a novel method to remove outliers based on density estimation and it has been applied to real-world traffic data. By comparison with the conventional approach, the experimental results indicate that the proposed algorithm is capable of detecting and removing outliers effectively for the data that may follow different unknown distributions, and the processed data retain the original and significant characteristics possessed by the system.