A direct proof is given that for optimal linear filtering in additive white noise, the causal (filtering) minimum mean-square error (MMSE) is equal to the non-causal (smoothing) MMSE averaged over the signal-to-noise ratio.