Various digital techniques for estimating the autocorrelation function of a signal are analyzed, with emphasis on methods applicable to radio-astronomy and incoherent-scatter studies of the ionosphere. The importance of coarse (one- or two-bit) quantization is stressed, as is the effect of oversampling, the use of a sampling interval less than the inverse of twice the signal bandwidth. For applications such as incoherent scatter in which the signal strength varies with range (i.e., time), multibit by one-bit multiplication combined with oversampling is probably the optimum strategy, although it requires 32% more integration time than the full multibit by multibit technique. For very high speed sampling of a signal of which the mean amplitude is constant, a three-level by three-level technique offers many advantages. It is nearly as simple to implement as conventional one-bit correlation but requires considerably less integration time to achieve equal statistical accuracy (only 26% more than full multibit correlation, when sampled at twice the Nyquist rate, as compared to 146% for a one-bit correlator with sampling at only the Nyquist rate).