Abstract: A novel window based data mining algorithm was proposed and used to mine the defined multivariate driving preference. Specifically, a model describing paired cost aspects to reflect driving preference was defined, a data mining algorithm to estimate the multivariate driving preference distribution was proposed, and ellipse to optimize the data mining algorithm was introduced. The results showed that the method in this paper could discover the defined multivariate driving preference and the proposed algorithm was effective, fast and had good scalability.