An inverse method to directly optimize the electrode configuration (positions, sizes, and driving voltages) for radio frequency (RF) capacitive hyperthermia was proposed. The main algorithm, based on the two-dimensional finite element method (2-D-FEM) solution of Laplace and bio-heat transfer equations, iteratively modified the individual boundary potentials around an object thereby making a calculated temperature distribution approach a target temperature distribution. A penalty function governed continuity and smoothness among the boundary potentials so that the optimized boundary potentials became attainable for two plate electrodes. Case simulations demonstrated the viability of the algorithm. For instance, in a computed tomography (CT)-based human abdomen model which had deep-and shallow-seated tumors, the optimized electrodes produced a temperature distribution suitable for heating the tumors; the average temperature differences between the tumors and normal tissues were 3.5 degree C for the deep-seated and 7.6 degree C for the shallow-seated tumors within 600 s of heating. A drawback with the present algorithm is that the choice of penalty coefficient and modification of the boundary potentials to coincide with the use of two plate electrodes are carried out manually. These procedures would be automated.