We present a method for the iterative refinement of triangulations. Given a coarse triangulation of the compact domain of a bivariate function, we present a refinement strategy bsed on approximation error. The triangulation is used to compute a best linear spline approximation, using the term best approxiamation in an integral least squares sense. We improve an approximation by identifying the triangle with largest error and refine the triangulation by bisecting this triangle.