Abstract: It is acknowledged by many AI researchers that machine learning is a promising appro-ach to the automatic knowledge acquisition of expert systems.In this paper we introduce our endeavorto apply inductive learning techniques to knowledge acquisition which resulted in a learning-basedknowledge acquisition tool GPMIL.Our research focuses on the following aspects:1.Synthetic inductivemethod for constructing a complete knowledge base;2.Inductive learning based on domain backgro-und knowledge;3.Learning algorithm with flexible and efficient search strategy to cope with largedescription space.An experimental application result is also presented.