A very important problem in the analysis of biological data sequences is the detection of oscillations in the presence of random variations (noise). If the oscillations are not stationary, i.e., if they drift in frequency and amplitude, or occur in bursts, traditional analysis techniques utilizing the power spectrum or its time-domain equivalent, the autocorrelation function, can be both misleading and insensitive. Temporal filtering by a comb or set of band-pass filters is very effective for identifying and describing nonstationary oscillations. The basic procedures for interpreting the output of a comb filter are presented, illustrated by examples using predefined input test sequences.