This paper introduces the causal reconstruction task-the task of reading a causal description of a physical system, forming an internal model of the specified behavior, and answering questions demonstrating comprehension and reasoning on the basis of the input description. A representation called transition space is introduced, in which events are depicted as path fragments in a space of "transitions," or complexes of changes in the attributes of participating objects. By identifying partial matches between the transition space representations of events, a program called PATHFINDER is able to perform causal reconstruction on short causal descriptions presented in simplified English. Simple transformations applied to event representations prior to matching enable the program to bridge discontinuities arising from the writer’s use of analogy or abstraction. The operation of PATHFINDER is illustrated in the context of a simple causal description extracted from the Encyclopedia Americana, involving exposure of film in a camera.