We present a cone-beam image reconstruction algorithm for helical CT scanning with a tilted gantry and N-PI data acquisition. When the gantry is tilted, the effective source trajectory in the patient's reference frame lies on an elliptical cylinder, rather than on a circular cylinder as in the standard helical scanning mode. The aim of this work is to provide a means of reconstructing an image object directly from cone-beam projection data without transforming the image object into a virtual object and without rebinning projection data acquired for a real object into the projection data of the virtual object. This task has been accomplished by the application of an exact reconstruction algorithm, which utilizes an important geometrical property of the elliptical helical trajectory: the existence of generalized N-PI lines for a given image point. Based on this property, a mathematically exact image reconstruction scheme via filtering the backprojection image of differentiated projection data (FBPD) is applied to solve the reconstruction problem. Due to the gantry tilt, the required detector size is different from that of the standard helical trajectory (nontilted). A systematic analysis of the required detector size is presented. For an N-PI data acquisition scheme, an image may be reconstructed using data from an N-PI window, an (N-2)-PI window, and so on. Although the images reconstructed using an N-PI (N>1) window are noisier than the images reconstructed from a 1-PI window, a weighted-average scheme over reconstructed images is presented to generate a final image with significantly lower noise variance than that in the 1-PI data acquisition scheme. The image reconstruction algorithm was numerically validated using a mathematical phantom.