In practice cone beam CT reconstruction is often done using the approximate Feldkamp-type algorithms because of their flexibility and simplicity; however, when it comes to a 64- and higher detector row CT, they produce cone beam artifacts, and for 256+ detector-row CT image quality becomes unacceptable. The exact algorithms, on the other hand, produce artifact-free images but are much more computationally demanding and not very flexible in controlling noise/resolution trade-off. We propose a cone beam algorithm that can be easily adjusted depending on the cone angle. We tested it on 64-slice real data and 320-slice simulated data. For 64-slice CT it outperforms helical Feldkamp in both image quality and volume reconstruction speed; using the same redundancy weighting it reduces the cone beam artifact and improves noise uniformity. Because weighting is done after the convolution step, we can use the n-PI Tam-Danielson window, an exact cone beam weighting. We show that the weighting is more important than the filtering direction for image quality. The proposed method has good image quality without rebinning to tangential filtering lines. In evaluation we compare the proposed algorithm to helical Feldkamp and the 1-PI and 3-PI Katsevich algorithms.