This paper describes a new method of model-based diagnosis that employs single fault process assumption. This assumption supposes that at most one fault process, which is a causal chain of physical phenomena that does not occur in normal states, will happen to an object machine at one time. By introducing this assumption, causally related multiple faults and the faults that cause structural changes in a machine are detected. Knowledge for diagnosis that is described based on qualitative physics is divided into two kinds; one is the object model that represents structure and characteristics of an object machine and the other is knowledge about principles and laws. In order to diagnose causally related multiple faults and the faults that cause structural changes, knowledge about physical phenomena is described in principles and laws. The algorithm of diagnosis consists of fault search and fault simulation. The former derives candidates of fault processes. The latter simulates behavior of the object machine in which a fault process occurs based on qualitative physics. By comparing simulated fault models with the object machine, the likeliest fault process is identified. An experimental diagnosis system that succeeded in diagnosing about some object machines is also described in this paper.