We introduce a policy-based slicing of database application. The policies are defined on data, that is kept in a backend database. The majority of the proposed slicing approaches are syntax-based and they do not consider the properties of variables and database-attributes. Therefore the existing approaches unable to compute precise slicing result. In this work, we refine (by removing false alarms) dependency graph using semantics-based technique. We formalize the data dependencies among various statements using semantics rule based approach and perform emptiness checking to identify the false alarm. This refined graph yields precise slicing results w.r.t the policies of a database application.