The　key　of　using　genetic　algorithm　to　mine　first-order　rules　is　how　to　precisely　evaluate　the　quality　of　first-order　rules.　By　adopting　the　concept　of　binding　and　information　theory,　a　new　fitness　function　based　on　information　gain　is　proposed.　The　new　fitness　function　not　only　measures　the　quality　of　first-order　rules　precisely　but　also　solves　the　equivalence　class　problem,　which　exists　in　the　common　evaluation　criteria　based　on　the　number　of　examples　covered　by　rules.　©　Springer-Verlag　Berlin　Heidelberg　2003.