In the magnetotelluric method, measurements of the components of the horizontal electric field are obtained by recording the voltage between grounded electrodes and dividing by their separation distances. This procedure gives the true value of the electric field only if it is uniform between the electrodes. In regions of near-surface inhomogeneity this condition is not fulfilled, and in extreme cases each electrode may be in contact with surface material of different resistivity. It is therefore suggested that voltages, rather than electric fields, should be computed in the two-dimensional modelling of such regions in the B-polarization mode, and that magnetotelluric impedance calculations for comparison with real data should be based on voltages. A method for modifying an existing finite difference program is described, and sample calculations of voltage differences in the control model of Weaver, LeQuang and Fischer are checked against the exact analytic results that can be obtained. Finally, real data obtained over the Gloucester Fault in Canada are compared with results given by finite difference modelling based both on voltage calculations and on the more conventional electric field calculations.