To determine the regional shear wave velocities in the lithosphere it is frequently necessary to use two-station dispersion curves. We investigate the influence of non-plane wave energy on such dispersion curves, and compare them to two other better known sources of error: deviations from great-circle paths and uncorrelated noise. To study the influence of non-plane waves created from distant heterogeneities we create complex wavefields in a laterally homogeneous medium by adding interfering waves to a main plane wave. We then calculate the apparent phase velocity between two seismic stations located 100–400 km apart. Using realistic values for the sources of error, we conclude that the contribution of each is similar for 200-km-long profiles. Our conclusions on non-plane waves are made under the assumption that non-plane energy from distant heterogeneities varies randomly with hypocentre location. If this is correct, then only five to 10 events with different hypocentres are required to obtain a stable dispersion curve with less than 1 per cent error. The influence of uncorrelated noise and non-plane waves diminish for longer profiles, while the errors due to great-circle deviations are independent of profile length and systematically bias the dispersion curve to higher velocities. We recommend the inclusion of some off-profile broad-band stations for surface wave studies on a regional scale, because such supplementary stations make it possible to apply first-order corrections for off great-circle propagation. The strong influence from the different sources of error, combined with our lack of precise knowledge of the nature and amplitude of non-plane energy, implies that the interpretation of two-station measurements should be restricted to major changes in regional earth structure.