Two major earthquakes, magnitude 7.0 in 1918 and magnitude 7.3 in 1946, have occurred this century in the central region of Vancouver Island, British Columbia, Canada. Levelling data in the region indicate relative uplift of 4 mm/year from 1977 to 1984, followed by subsidence at approximately the same rate over the next 2 years. In response to the observed elevation changes, a program was initiated to investigate if temporal changes in the geoelectrical conductivity might be associated with earthquake occurrence. Beginning in 1986, magnetotelluric (MT) data have been measured annually at a number of sites on central Vancouver Island to monitor the long-term variability of the conductivity of the crust and upper mantle in the region. Robust processing techniques now used in the analysis of MT data enhance the possibility of detecting changes in the conductivity.Past studies involving the monitoring of MT stations have considered temporal change only in terms of the measured responses. However, formulating the inverse problem of constructing conductivity–depth models that vary minimally from year to year allows quantitative investigation of the changes required in the models to accommodate the yearly variations in the data. This provides a method of evaluating the processes and depths involved in observed changes in the data. Our modelling study indicates a small but systematic yearly decrease in conductivity from 1987 to 1990 localized in a conductive zone overlying the subducting Juan de Fuca Plate.