In the field of legacy software maintenance, there unexpectedly arises a large number of implicit coding rules, which we regard as a cancer in software evolution. Since such rules are usually undocumented and each of them is recognized only by a few members in a maintenance team, a person who is not aware of a rule often violates it while doing various maintenance activities such as adding a new functionality or repairing faults. The problem here is not only such a violation introduces a new fault but also the same kind of fault will be generated again and again in the future by different maintainers. This paper proposes a method for detecting code fragments that violate implicit coding rules. In the method, an expert maintainer, firstly, investigates the cause of each failure, described in the past failure reports, and identifies all the implicit coding rules that lie behind the faults. Then, the code patterns violating the rules (which we call "faulty code patterns") are described in a pattern description language. Finally, the potential faulty code fragments are automatically detected by a pattern matching technique. The result of a case study with large legacy software showed that 32.7% of the failures, which have been reported during a maintenance process, were due to the violation of implicit coding rules. Moreover, 152 faults existed in 772 code fragments detected by the prototype matching system, while 111 of them were not reported.