We present a deconvolution for automated magnetic interpretation based on Werner's (1953) simplified thin-dike assumption which leads to the linearization of complex nonlinear magnetic problems. The horizontal gradient of the total field caused by the edge of a thick interface body is equivalent to the total field from a thin dike. Statistical decision, numerical iteration, and a seven-point operator are used to improve approximations of susceptibility, dip, depth, and horizontal location of the source. Marquardt's nonlinear least-squares method for inverse modeling is then used to refine automatically the first approximation provided by the deconvolution. Synthetic and real total-field data are used to demonstrate the process.--Modified journal abstract.