This　study　provides　investigations　on　whether　the　fracture　toughness　can　be　accurately　predicted　from　mini-sized　uniaxial　tensile　specimens　with　existing　global　and　local　approaches.　The　critical　toughness　model,　based　on　classical　fracture　mechanics,　and　the　energy　release　rate　(ERR)　model,　based　on　damage　mechanics,　were　used　as　the　two　representative　global　approaches.　While　the　finite　element　simulated　compact　tension　(CT)　specimen　implemented　with　Gurson-Tvergaard-Needleman　model　was　used　as　the　local　approach.　Specimen　size　effect　was　judged　through　the　viewpoints　of　stress-strain　relationship　and　damage　development.　It　was　proved　that　the　specimen　size　has　no　significant　influence　on　stress-strain　relationship,　and　has　little　influence　on　damage　developments　when　the　thickness　reaches　1.0　mm.　Fracture　toughness　predictions　were　conducted　on　four　metals,　including　two　steels　(SA508Gr.3Cl.1　and　18MnMoNbR),　one　aluminum　alloy　(6061-T651),　and　one　titanium　alloy　(TC4).　For　TC4　exhibiting　brittle　cleavage　features　on　the　fracture　surface　of　CT　specimens,　neither　the　global　nor　the　local　approach　can　accurately　predict　its　fracture　toughness.　For　6061-T651　exhibiting　both　ductile　and　brittle　features　on　the　fracture　surface　of　CT　specimens,　there　exists　significant　macroscopic　crack　propagation　at　the　moment　of　conventional　fracture　toughness　determination.　The　existence　of　significant　macroscopic　crack　is　inconsistent　with　the　＂equivalent　crack＂　assumption　in　the　ERR　model,　leading　to　a　failure　in　its　fracture　toughness　prediction.　For　two　steels　following　the　＂equivalent　crack＂　assumption,　the　ERR　model　and　the　local　approach　yield　more　accurate　predictions　(with　the　maximum　error　around　10　%)　than　their　critical　toughness　counterpart　(with　the　maximum　error　around　30　%).