The standard Werner deconvolution method is extended to include the effects of variable sensor altitude but this leads to a deconvolution algorithm that is unstable for slowly changing flight height. By expressing the sensor altitude as a linear function of horizontal position (within a specified window), we show that the numerical instability can be avoided. The subsequent selection and averaging of the raw solutions is controlled by three parameters that can be adjusted to specific survey data characteristics. Results for an aeromagnetic survey over Vancouver Island, British Columbia show that, in comparison with the variable altitude approach, the standard Werner method produces unacceptable errors when applied to variable altitude data.