We investigate the way data are used in the algorithm proposed by Kudo and Saito for the exact reconstruction of long objects from axially truncated cone-beam projections. Specifically, we show that the algorithm wastes a large part of the data. To overcome the problem, we propose to use a vertex path consisting of two crossing ellipses, for which we devised a new reconstruction algorithm, called the cross algorithm, which does not waste data and is still suitable to exactly handle axial truncation. Results of reconstruction are presented on simulated data and real data from an experimental scanner.