The performance of three post-stack seismic inversion techniques has been evaluated using synthetic and VSP data. The application of classical recursive inversion to simple synthetic models demonstrates the importance of low frequencies in trace inversion. This is consistent with the strongly low-pass response of the inversion transfer function. Impedance profiles generated from band-limited seismic traces contain only “depth-positional” information. An accurate impedance profile can only be obtained if independent, reliable low-frequency information is incorporated following inversion. Autoregressive (AR) spectral extension is evaluated as a representative of sparse-spike inversion techniques. Synthetic examples suggest excellent performance on sparse-spike models, but performance deteriorates as the reflectivity series becomes more dense. Poor performance on real VSP data is a further indication of the failure of the AR technique when fundamental assumptions are not satisfied. Of the three methods investigated, model-based linear inversion is preferred in terms of accuracy, robustness and logistics. In contrast to previously published modelling algorithms, no geological interpretation is required prior to inversion. Instead, the inverted band-limited trace is used to initiate the inversion scheme. Critical low-frequency control is achieved by incorporation of RMS velocity constraints. The robustness of the model-based inversion technique when applied to VSP data suggests that it is possible to recover accurate impedance information beyond the bottom of the borehole.