Explanation—based learning(EBL)methods learn from single training ex-ample．The learning result often bears the example's own speciality．The knowledge re-finement algorithm can rectify the speciality in EBL，but with rather low utility．This pa-per combines EBG and refinement algorithm，gives an incremental learning algorlthm——EBG—plus，which can take advantage of many examples．While maintaining high utility，the authors get better result as new instances are met．By the way，the quality of domain knowledge can be automatically improved．