The　paper　proposes　an　extended　alias　analysis　method　for　improving　the　binding　time　analysis　(BTA)　precision　in　the　implementation　of　automatic　program　specialization　for　Java　programs.　In　the　object　alias　analysis,　a　class　specialization　model　is　generated　by　tracing　uses　of　reference　variables　across　classes　and　methods　in　Java　to　represent　the　BTA　features　of　multi-dispatch　and　complex　data　structures　such　as　member　reference　variables.　The　poly-variant　specialization　approach　can　improve　accuracy　of　partial　evaluation　on　Java　classes　and　their　methods,　and　has　been　applied　to　implementation　of　the　binding　time　analysis　in　a　partial　evaluator　for　Java　programs.