Traditional Chinese Medicine (TCM) has been a successful medical science with the clinical practice of thousands of years.Chinese medical formula (CMF) is one of the main therapies in TCM.Inducing the practical CMF knowledge(like core herb combinations and modifications) from the daily clinical treatment is an important step for young doctor learning and efficient formula formation.This paper proposed a network and graph based data mining method to discover the hierarchical core herb combination structures from the large amount of formulae.Recently, discovery of the community and hierarchical organization structures from large complex network (e.g.scale-free network) has been the foci of complex network researches.However,most of the community discovery methods are to get the clustering structure of the whole network scale (with all nodes).They can not extract the hierarchical core structure with exemplar characteristics from the whole network.Because of the complex conditions of the treatment and patients, distilling and induction of core knowledge with common disease conditions from the large amount of formulae is difficult and significant for formula knowledge discovery.We have conducted several empirical data analysis tasks on the TCM clinical data from the outpatient treatments of famous TCM physicians.The results show that we can get the clinically interesting and important herb combination knowledge from TCM clinical prescriptions by network based method.