To build model-based systems capable of emulating the scientist's or engineer's way of reasoning about a given physical domain requires methods for automating the formulation or selection of a model which adequately captures the knowledge needed for solving a specific problem. To find and exploit such models requires the use and integration of different kinds of knowledge, formalisms and methods. This paper describes a system which aims at reasoning automatically about visco-elastic materials from a mechanical point of view. It integrates both domain-specific and domain-independent knowledge in order to classify and analyse the mechanical behaviour of materials. The classification task is based on qualitative knowledge, whereas the analysis of a material is performed at a quantitative level and is based on numerical simulation. The key ideas of the work are to automatically generate a library of models of ideal materials and their corresponding qualitative responses to standard experiments; to classify an actual material by selecting from within the library a class of models whose simulated qualitative behaviours towards standard loads match the observed behaviours; to identify a quantitative model of the material, and then to analyse the material by simulating its behaviour on any load. Each model in the library is automatically generated in two different forms; at the lowest level, as a symbolic description and, at a mathematical level, as an ordinary differential equation. This paper mainly concentrates on the methods and algorithms of model generation and qualitative simulation.