We propose an inverse optimizing method, used in rf-capacitive hyperthermia to obtain a certain target tumour temperature and to avoid the over-heating of normal tissues. Based on the finite-element-method solutions of the Laplace electric potential equation and Pennes bio-heat transfer equation, the heating conditions are modified by the genetic algorithms recursively to minimize the objective function for searching an optimum physical configuration. With a simplified human tissue model extracted from an x-ray computed tomography slice which has deep- and shallow-seated tumours, satisfactory simulation results are obtained both on bi-plate and three-plate systems, e.g. the tumour temperature is higher than 43℃ and the temperature of normal tissues is lower than 39℃ . It is suggested that the proposed algorithm is suitable for both shallow and deep seated cancer oncology.