General System Theory (GST) sets its goal as unification of science, and its subject matter as formulation of general system principles, or principles applicable to all systems. Unfortunately, no such principles have been formulated to date. This paper demonstrates that GST is incapable of formulating system principles due to its naturalistic methods and realist ontology that represents system as a real object, or thing. This paper suggests an alternative approach to development of system principles based on the assumption that the purpose of the system approach is not unification of science but dealing with complexity, and the corresponding method is analysis–synthesis. According to this approach, reality is neither systemic nor non-systemic, and only our methods of treating real objects and corresponding representations of them are either systemic or not. The paper asserts that system science cannot study real systems, because they do not exist. Instead it should study history of system reasoning as the process of development of system principles. As a preliminary step, by analyzing classical definitions of system, this paper identifies the principal system concepts and categorical oppositions that can be used as an initial framework for empirical studies of system reasoning. It also outlines methodology for such studies. Copyright © 2004 John Wiley & Sons, Ltd.