Value dependence is an emerging topic in the research of database inference area. Firstly, aiming at the characteristics of value inference and requirement of formal concept models, formal context and concept lattice are presented. Then, formal concept models are investigated, and the formal definition of value dependence and formal concept models are proposed. Next, the relationships between intent reduction of concept lattice and value dependence are analyzed. Two new concepts, inference dependence and α maximum inference dependence, are proposed by introducing the security sensitivity level of the property into value dependence. Furthermore, aiming at the generation rule of α maximum inference dependence, the properties of intent reduction set of formal concept lattice are studied. Specifically, it is proved that a perfect, non-redundant α maximum inference dependence set can be deduced from the intent reduction of formal concept lattice. Finally, an algorithm based on intent reduction of mining the full inference dependence set in database is proposed, and a case study is implemented to prove its effectiveness. In a global view, the study of inference dependence, which is one of the most important attribute dependencies in relational database, shows great insights on the detecting and eliminating database inference channel.