Financial　early　warning　(FEW)　models　aim　to　help　companies　recognize　possible　financial　crises　and　reduce　financial　risks　through　generating　FEW　messages.　The　current　FEW　models　are　mainly　constructed　by　a　set　of　financial　indicators　and　the　predictive　accuracy　has　only　been　verified　by　these　financial　indicators　rather　than　non-financial　indicators.　The　issue　in　such　a　situation　is　that　these　financial　indicators　can　be　controlled　or　manipulated　by　related　senior　managerial　personnel　of　companies,　and　therefore,　using　only　financial　indicators　to　verify　FEW　models　cannot　ensure　the　reliability　of　predictive　accuracy　of　the　models.　To　handle　this　issue,　this　paper　develops　a　new　FEW　logit　model　which　has　better　predictive　accuracy　than　existing　ones.　More　importantly,　we　propose　a　new　approach　which　verifies　the　predictive　accuracy　of　the　logit　model　by　using　non-financial　efficiency　indicators　of　data　envelopment　analysis.　An　empirical　study　on　Chinese　company　datasets　revealed　that　the　accuracy　rates　of　predictions　of　the　proposed　model,　for　in-sample　and　out-of-sample　companies,　are　97.1%　and　94.1%　respectively,　higher　than　existing　results.　Using　non-financial　efficiency　indicators,　the　verification　rates　for　the　prediction　results　of　the　logit　model　for　in-sample　and　out-of-sample　companies　are　95.8%　and　96.2%,　respectively.　The　findings　show　that　the　proposed　FEW　logit　model　has　improved　the　accuracy　of　prediction　and　stability;　the　approach　which　uses　non-financial　efficiency　indicators　to　verify　the　results　of　FEW　logit　model　has　significantly　ensured　the　reliability　of　the　FEW　models.　(C)　2014　Elsevier　B.V.　All　rights　reserved.