Business rules are a set of conditional operations attached to a given data result. On legacy systems, it is very difficult to extract business rules because of the inconsistency of documentation. Some techniques have been presented for extracting business rules from legacy systems. But usefulness of these methods is limited when they are applied to large complex legacy systems. Generally, large legacy systems involve large amount of code, domain variables, synonym variables and business rules, which make it more difficult to extract business rules. This paper proposes a framework, which offers distinct advantages over normal extraction solutions for large legacy systems. This framework consists of five steps: slicing program, identifying domain variables, data analysis, presenting business rules, and business validation. It has been applied to a large complex financial legacy system which has proved to be successful.