This paper reports an ongoing feasibility study in industrial X-ray tomography. Full volume scanning of logs may have a large potential in terms of increased sales value. On the other hand, there is no technology commercially available today which can meet the requirements for speed. In conventional medical 3D-tomography, the source-detector system rotates around a slowly translating patient. We do not believe such an arrangement is able to reach our speed requirements. Instead, we have been simulating an arrangement with two fixed source-detector systems rotated 90° relative to each other. Each system consists of an X-ray source and a 2D detector. The logs are to be translated through this arrangement at relatively high speed (2-3 m/s) lengthwise on a conveyor belt, while cone-beam projections are acquired by each of the source-detector systems. The 2D-detectors are composed of a number of 1D-detectors side by side. For cost reasons we will try to limit the number of rows to nine, forming a 2D-array of 256 x 9 detector elements. Due to the unconventional scanning geometry, we have developed a new reconstruction algorithm, which also takes special care to avoid many artifacts due to sparse and/or missing data. Experiments (still simulated), performed on voxelized CT data, indicate that the knots are reconstructed with sufficient accuracy to allow for quantitative optimization. Due to the missing data, however, heartwood can barely be distinguished from sapwood, but his fact does not seem to compromise the subsequent knot detection. Note that this paper only describes the reconstruction part of the simulated experiments. To be complete, the feasibility study should also include segmentation, which is outside the scope of this paper. © 2003 Elsevier Science B.V. All rights reserved.