On　construction　sites,　there　are　many　catastrophic　accidents　induced　by　human　error.　The　fuzzy　cognitive　reliability　and　error　analysis　method　(CREAM)　is　an　effective　method　for　assessing　human　error　hazards　in　such　a　context.　However,　deficiencies　regarding　the　aspects　of　reasonably　considering　the　input　weight,　evaluation　of　common　performance　conditions　(CPCs),　and　the　rule-based　approach　for　fuzzy　illation　are　frequently　encountered.　This　paper　provides　an　improved　weighted　fuzzy　CREAM　model.　First,　the　reasonable　weight　of　the　CPCs　is　obtained　by　using　multiple　correlation　analysis　and　evidence　theory.　Second,　the　specific　evaluation　rules　for　CPCs　are　established　based　on　Chinese　codes,　and　the　membership　function　for　each　CPC　level　is　obtained　by　a　statistical　method.　The　discrete　basic　diagram　for　the　control　mode　is　then　transformed　into　a　continuous　fuzzy　rule-based　model.　On　that　basis,　the　rationality　and　reliability　of　the　model　are　verified　by　four　axioms.　The　method　that　introduces　the　weights　into　the　calculation　of　the　expected　effects　of　CPCs　and　the　reasoning　of　human　error　probability　(HEP)　can　help　evaluate　the　effectiveness　of　CPCs　on　human　performance.　The　model　is　sensitive　to　the　minor　alterations　of　CPC　scores　and　weights;　the　sufficiency　of　the　data　utilization　and　the　solution　domain　are　also　verified.　An　exemplified　application　shows　the　rationality　and　validity　of　the　new　weighted　fuzzy　CREAM　for　shield　tunneling　human　reliability　analysis.　This　model　also　provides　essential　HEP　data　that　can　be　used　in　hazard　analysis　in　other　engineering　fields.