We have recently introduced a novel approach to sonar based object recognition for robotics. The sonar recognition system, consisting of a Polaroid sonar coupled with art A/D data acquisition board in a LINUX-based PC, uses a fuzzy ARTMAP neural network classification system to recognize objects at varying distances. In this article we report additional results in which we test systematically recognition performance using different kinds of information from the sonar echo, and different object configurations. The results strengthen our claims that sonar can be used as a viable system for real-time object recognition in robotics and other application domains.