The improvement of spatial dose distribution is one of the means to make feasible the local control of malignancies, which is the goal of radiotherapy. We have modified the inverse planning introduced by Brahme in the late 1980's and developed a new high-performance calculation method, confirming its validity in a case of head and neck tumor. The employed prescription of dose distribution conforms to DICOM-RT, and the object function comprises degrees of dose dissatisfaction regulating PTV and OARs. The treatment parameters are defined as the intensity modulated beam weight distribution for all portal beams. A multiplestep gradient method is used to optimize dose distribution. Graphic display for each term of the object function allows visual estimation of its convergency. Initial values for each local search are chosen from a solution of the prior search, while slightly tuning the prior prescription. We have confirmed the availability of this method by applying it to a case of oropharynx carcinoma, in which we used fixed five ports on a 2D image. Convergence was met after twenty iterations for the first search, while the second search required only five iterations. The DVH analysis revealed that OAR exposure can be reduced if we allow increased dose non-uniformity in PTV. Inverse planning has been clinically applied chiefly with a 2.5 dimensional technique to prostate carcinoma as well as head and neck tumors. We believe that three dimensional implementation with high degrees of beam direction freedom is required so as to be applied to such cases as hepatocholangio-pancreatic malignancies where dose optimization is expected to accompany greater difficulties. Global search strategy seems to run the risk of vast calculation time. The advantages of the proposed multiple-step gradient method with prescription tuning capability are fast calculation and reduced risk for local minimum problems. It is highly expected that 3D inverse planning being developed will become a competent rival to proton therapy.