A time averaging filter for channel estimation is described. Different time averaging is used for different taps or tones of a communication system. Two methods are suggested for determining the time-averaging filter for channel estimation for the communication systems without prior knowledge of SNR, channel Doppler and fading statistics. The first method includes a bank of different filters, each tuned for optimality in a certain range of SNR and Doppler. For each tone in the frequency domain (or tap in the time domain), the first method selects a filter among the set of filters that minimizes a certain computed metric. The second method directly computes, at each tone (or tap), the “best” filter in a least-squares sense, using empirically determined second-order statistics.