A computer simulation is used to investigate the relationship between skin impedance and image artefacts in electrical impedance tomography. Sets of electrode impedance are generated with a pseudo-random distribution and used to introduce errors in boundary voltage measurements. To simplify the analysis, the non-idealities in the current injection circuit are replaced by a fixed common-mode error term. The boundary voltages are reconstructed into images and inspected. Where the simulated skin impedance remains constant between measurements, large impedances (> 2k omega) do not cause significant degradation of the image. Where the skin impedances 'drift' between measurements, a drift of 5% from a starting impedance of 100 omega is sufficient to cause significant image distortion. If the skin impedances vary randomly between measurements, they have to be less than 10 omega to allow satisfactory images. Skin impedances are typically 100-200 omega at 50 kHz on unprepared skin. These values are sufficient to cause image distortion if they drift over time. It is concluded that the patient's skin should be abraded to reduce impedance, and measurements should be avoided in the first 10 min after electrode placement.