This paper studies a design method of decentralized signal detection system which consists of adaptive fuzzied local-detectors and a data fusion rule of on-line self-learning weights. The local--detectorS for inaccurate signal parameters are modeled by means of fuzzy sets which can be adapted to change of the inaccurate signal paraxneteres. The data fusion center Where the optimal decision rules are used as objective function can learn the local decision weights on-line. The robustness of the fuzzied local-detectors and the adaptability of the self-learned fusion rule make it true that the detection performance of the decentralized detection system is improved under uncertainty and this system can also process the decentralized signal detection with a unknown parameter of unknown distribution or non-random unknown parameter.