A transformer internal composite defect fuzzy diagnosis method based on gas dissolved in oil, comprising: a step of acquiring monitoring data of volume concentrations of five types of monitored feature gas; a step of determining ratio codes; a step of modifying a three-ratio method; a step of fuzzifying a boundary range; a step of calculating probabilities of the ratio codes; a step of calculating a probability of occurrence of each defect fault; and finally obtaining a fault type of a transformer. The method has the beneficial effects that: the method is simple and easy to achieve, and particularly suitable for being applied to an on-line transformer state monitoring system; based on a concept of fuzzy logic, diagnosis of composite defects of the transformer under a complicated state and evaluation of the degree of severity can be achieved, and the problem of sudden change caused by criterion boundary absolutisation can be effectively avoided; and multi-feature information such as an attention value and a ratio of the gas dissolved in the oil are merged and analysed, thereby effectively improving the diagnosis reliability.