We present a novel reconstruction algorithm for high-transaxial-resolution cone-beam computed tomography (CT). Two ideas are combined to achieve this: First, the ray-offset technique, which is well-known for two-dimensional reconstruction, is adapted to the case of cone-beam CT. Instead of interleaving parallel re-binned projections from opposite directions during pre-processing before radial re-binning and filtering, the parallel re-binned projections are interleaved with virtual projections, which contain only zeros. These new projections are processed further as usual. The actual combination of opposite measured projections is moved into the back-projection step of the reconstruction. The second idea is the so-called frequency-split technique, which is designed to minimize cone-beam artifacts. Observing that cone beam artifacts contain mainly low-frequency components, the reconstructed image is a combination of two independent reconstructions with differently modified reconstruction filters and different amounts of redundant data. A high-frequency image is reconstructed with the ray offset technique using all the redundant data, and a high-pass filter leading to a suppression of low-frequency artifacts, while the low-frequency components of the image are reconstructed by using little redundant data only. The pro- posed technique is applied to the WEDGE algorithm and helical acquisition.