To enable simplification of reconstruction processing for fractional cone beam data while allowing enhanced efficiency in the reconstruction processing.

A method of reconstructing an image from fractional cone beam data contains the collection of the partial cone beam data in a two-dimensional array. When the apex advances along a curve, the collected data coincides with the beam of the radiation diverged two-dimensionally from the common apex. The elements of the data present a line integral of an object taken along the beams to be reconstructed. Thereafter, the data are previously weighted 52 and a local coordinate system is defined. The one-dimensional convolution 54 of the data previously weighted is calculated by the local coordinate system along the direction in which the apex contacts the advancing curve. Finally, the data thus convoluted is weighted while being back projected into an image memory 60 three-dimensionally. In the ideal embodiment, the curve prescribes a spiral path and the kernel for the one-dimensional convolution of the data pre-weighted is a ramp convolver.