An optimal one-weight "filter" is presented for the purpose of improving the signal-to-noise ratio of averaged ECG recordings in the analysis of late potentials. Based on a simple statistical model, the filter is estimated from the ensemble correlation of available beats. The correlation estimator is found by a maximum likelihood procedure in which the observed signal is assumed to have a Gaussian distribution. The performance of the optimal filter is studied in relation to an ensemble with individual or subaveraged beats.