A good model is essential in simulating and analysing a system satisfactorily. This paper proposes a new computer-aided modelling methodology to construct parsimonious models for dynamic physical systems. A set of domain-independent modelling principles and domain-dependent modelling constraints has been formulated to guide parsimonious model construction. Models and the knowledge base for the enhancement of models have been described by bond graphs. After a parsimonious model is generated, two types of qualitative analysis are conducted; one is to derive unknown states from known states and the other is to predict the effects of parameter changes on the whole system. This methodology is only applicable when the structure of a dynamic system is known beforehand. A detailed case study is presented to demonstrate this approach.