For nonlinear transient heat transfer system, a fuzzy adaptive predictive inverse method (FAPIM) is proposed to inverse transient boundary heat flux. The influence relationship matrix is utilized to establish time-varying linear prediction model of the temperatures at measurement point. Then, the predictive and measurement temperatures are used to inverse the heat flux at current moment by rolling optimization. A decentralized fuzzy inference (DFI) mechanism is established. The deviation vector of the predictive temperature is adopted to conduct decentralized inference by a set of fuzzy inference units, and then, the influence relationship matrix is updated online to guarantee the adaptive ability of the prediction model by weighting fuzzy inference components. FAPIM is utilized to inverse the unknown heat flux of a heat transfer system with temperature-dependent thermal properties, which has shown that the inverse method has better adaptive ability for the inverse problems of nonlinear heat transfer system.