Lane detection and classification is a key research of Lane Change Assist System (LCAS). It is a hard problem that how to classify different typesSof lane how to the lanes accurately. In this paper a robust lane detection method was proposed based on steerable filter. Firstly road surface in several windows is sampled in YCbCr color space and a gray Gauss distribution model is built for road surface segmentation. A steerable filter is then designed for lane edge filtering within road surface region. The direction angle θ of the filter could be initialized by the histogram of gradient direction. Next a gray accumulation strategy is set up to reduce gray shift of lane region caused by illumination change. Dynamic lane ROI is designed by Hough straight line model. FinallySlane classification method is proposed for distinguishing single dashed lane and solid lane based on the gray mean statistics of ROI within specific time window. Highway experimental results demonstrated that the classification accuracy rates of single dashed lane and solid lane reach 88.5% and 90.3%, respectively. The proposed algorithm is robust to interference from road environment and illumination. This research isSsignificant for both the further optimization of lane change warning strategy and the enhancement of safety and intelligence level of LCAS.