A crucial step in obtaining high-order accurate steady-state solutions to the Euler and Navier-Stokes equations is the high-order accurate reconstruction of the solution from cell-averaged values. Only after this reconstruction has been completed can the ux integral around a control volume be accurately assessed. In this work, a new reconstruction scheme is presented that is conservative, uniformly accurate with no overshoots, easy to implement on arbitrary meshes, has good convergence properties, and is computationally e cient. The new scheme, called DD-L 2 , uses a data-dependent weighted least-squares reconstruction with a xed stencil. The weights are chosen to strongly emphasize smooth data in the reconstruction. Because DD-L 2 is designed in the framework of k-exact reconstruction, existing techniques for implementing such reconstructions on arbitrary meshes can be used. The new scheme satis es a relaxed version of the ENO criteria. Local accuracy of the reconstruction is optimal except for functions that are continuous but have discontinuous low-order derivatives. The total variation of the reconstruction is bounded by the total variation of the function to within O (x). The asymptotic behavior of the scheme in reconstructing smooth and piecewise smooth functions is demonstrated. DD-L 2 produces uniformly highorder accurate reconstructions, even in the presence of discontinuities. Two-dimensional ow solutions obtained using DD-L 2 reconstruction are compared with solutions using limited least-squares reconstruction. The solutions are virtually identical. The ab