Downwind computational boundaries in the numerical approximation of hyperbolic equations are in general not transparent, and they create spurious reflection. A useful measure of this is given by the energy (or usual sum of squares) of the reflected solution in response to an arbitrary solution which originates from within the computing domain. We prove, in that respect, a somewhat unexpected property: namely, that for those full-discretizations which are obtained by applying to a space-discretization of the equations an energy conservative discrete time-marching method, the energy reflected at the boundary is independent of the value of Δt, and is strictly equal to the energy reflected in the semidiscrete case. This is verified in numerical experiments. Optimal boundary equations may be defined in the semidiscrete case of those which maximize the rate of convergence to zero of the reflected energy when h-+0. A corollary of the preceding result is that those boundary equations remain optimal, in the same sense, when used in an energy conservative full discretization. Moreover, this convergence result continues to hold when a nonconservative but stable (i.e., dissipative) time-discretization is used. This is verified numerically with the first-order Adams predictor-corrector method. The results of this paper are derived with the mathematics of the simple three-point central difference discretization of spatial derivatives. Obvious generalizations to other cases are mentioned at the end of the paper.