An approach for 3D surface reconstruction from two uncalibrated views is proposed in this paper. The SUSAN corner detector is used for corners from points in edges, which are detected by Sobel edge operator. After corners matching between two images, the fundamental matrix is estimated to acquire 3D structure from matched corners by an improved weighted linear algorithm, which is based on the epipolar geometry and the absolute conic theory. The triangular textures are acquired by dividing 2D image. Then a realistic 3D surface model is built by mapping the triangular textures to the 3D structure surface. Experiments show that the reconstructed 3D surface models are satisfactory and photo-realistic.