Abstract: Takifugu rubripes belongs to the bony fishes,Tetraodontiformers,Tetraodontoidei,Tetraodontidae, Takifugu.It is distributed mainly in Japan of the western north Pacific,the Korean peninsula and China's coastal areas.Due to its appealing taste,rich nutrition,low fat content and numerous trace elements, Takifugu rubripes represents one of the fish species with high economic value.In recent years, Takifugu rubripes are farmed in large numbers in Dalian,Qinhuangdao,Tangshan,and Tianjin regions,and has become the main cultured species of puffer fishery in china.There existed large errors due to self-correlation between different phenotypic traits,non-linear relationship between some traits and body weight and the collinearity among independent variables,when the linear regression model was used to predict Takifugu rubripes weight.As a solution,a Takifugu rubripes weight prediction RBF neural network model,according to Artificial Neural Networks theory and Radial Basis Function model,was constructed with the phenotypic traits(including total length,body length,body depth,head length,length between eye and head,snout length,mouth width,eye diameter,space between eye and eye,caudal peduncle length,caudal peduncle depth,caudal peduncle breadth,body width,trunk length,trail length,body girth 1,body girth 2 and body weight) of 72Takifugu rubripes based on the nearest neighbor clustering algorithm,and the credibility of the neural network model constructed was tested by linear regression techniques.The results showed that the coefficient of determination R² of RBF neural network prediction model and the linear regression model for Takifugu rubripes weight were 0.992(approximately 1) and 0.949,respectively.Obviously,the coefficient of determination R² of RBF neural network prediction model was improved by 4.53% compared with the linear regression model.In addition,the collinearity diagnostics of linear regression,based on tolerance and variance inflation factor as well as maximum condition index and maximum variance proportions,indicated that there existed certain collinearity among independent variables and self-correlation between body girth 1,and body depth.The results suggested that the RBF neural network technique was an effective method to construct the prediction model of Takifugu rubripes,and the collinearity of the independent variables,in RBF neural network analysis,was eliminated and it has higher accuracy than linear regression prediction model.Weight prediction model based on radial basis function neural network provides a new method for accurate prediction of Takifugu rubripes weight.