Abstarct: In this thesis Data fusion in uncertainty condition defined as "combining several uncertainty sources to make an effective representation for human or machine by ability to handle uncertainty and inconsistency”. The most important aspect of this thesis is identifying and dealing with uncertainty and inconsistency in unknown systems. Most previous data fusion methods such as Kalman filtering are dependent on the system behavior. Such dependency does not allow us to easily deal with unknown systems.in systems with behavior model, data fusion can be applied on two sources. But in unknown systems data fusion needs many sources. The proposed algorithm consists of a new clustering technique, neural network and finally new update prediction rules for the predictor. In general, when the sources contain uncertainty and inconsistency, data fusion may fail. The proposed method in this thesis can recognize and then remove the inconsistent data points, thus it presents reliable results. The experimental results on both synthetic and real data (weather forecast) confirm the effectiveness of the proposed approach. Finally we examined our proposed method using Meteorological data. The experimental results show strength of the proposed method and its ability to cope with uncertain and inconsistent sources.