Different seismic pulse compression methods are evaluated. These include several algorithms for computing prediction error filters: Wiener filtering, Burg's method, the l 1 norm criterion, Kalman filtering, and two time-adaptive methods. Algorithms which do not assume a minimum-phase condition for the seismic wavelet include minimum entropy, homomorphic, and zero-phase deconvolution. The sensitivity of these algorithms is examined for various earth reflectivity functions, source waveforms, and signal distortions. The results indicate that standard Wiener predictive deconvolution is robust under a wide variety of input conditions. However, a substantial improvement in pulse compression can be obtained by the Burg algorithm under conditions of short data segments and by minimum entropy deconvolution for seismograms consisting of mixed-phase wavelets combined with sparse reflectivity series.