This paper presents how to detect planes from a set of corresponding points with a stereo camera system. We show that, in the presence of two cameras aligned to the same orientation, it is possible to compute the normal vector to a plane by using only three corresponding points whereas with traditional methods this computation requires at least four points and the decomposition of the homography matrix. We obtain the triangulation of the set of features and we compute for every triangle its normal vector. By grouping adjacent triangles with equal normal we are able to select all the points belonging to the same plane. The method does not require any information about the camera calibration or the camera relative translation. Simulation and experiments on real images validate the proposed algorithm.