The　predicted　mean　vote　(PMV)　and　its　several　revised　models　are　widely　used　for　the　prediction　of　thermal　comfort.　This　study　aims　to　assess　their　performances　using　the　Chinese　Thermal　Comfort　Database　(N　=　41977).　In　air-conditioned　buildings,　the　PMV　prediction　accuracy　(P)　and　the　mean　absolute　error　(MAE)　are　41.2　%　and　0.86,　respectively,　which　is　better　than　the　performance　in　free-running　buildings　(P　=　31.9　%,　MAE　=　1.09).　The　performance　of　the　PMV　model　is　also　tested　under　different　HVAC　modes,　climate　zones,　and　building　types.　The　prediction　accuracy　varies　but　does　not　exceed　60　%　for　all　subset　cases.　Three　typical　revised　models　(ePMV,　nPMV　and　aPMV)　considering　thermal　adaptation　show　better　accuracy　than　the　PMV,　but　the　improvements　are　still　limited　and　do　not　exceed　5　%.　It　appears　that　the　PMV　and　revised　models　are　reliable　under　thermal　neutrality　conditions,　while　their　accuracy　decreased　towards　the　ends　of　the　thermal　sensation　scale,　especially　on　the　cooler　side.　For　further　improvement　of　the　prediction　performance,　it　may　be　necessary　to　consider　the　effect　of　thermal　adaptation　in　parallel　with　other　approaches,　such　as　revising　the　PMV　core　structure　and　considering　individual　differences.　©　2022　Elsevier　B.V.