Abstract: This paper explores an induction-based discovery system which can get abstract concepts, structural models or mathematical expressions from the original numeric or symbolic data provided by experiments. They are concept learning system （CL）. analysing components of compound （AC）. constructing structural models （SM） and MP system which discovers the polynomial relations. such as Y=F（x₁, x₂, ……x_n）. The CASM is better than BACON system in the combination of qualitative and quantitative concepts and the discovery of multivariable empirical relations.