Abstract: Based on the compression experimental data of BFe10-1-1 alloy with continuous unidirectionally solidified columnar grains, a prediction model for the relation of true stress to temperature, strain rate and true strain was developed using an adaptive network based fuzzy inference system (ANFIS). The temperature at which the alloy was compressed was from 25 to 500℃ with the strain rate ranging from 0.01 to 10s^-1. Simulation results show that the mean percentage error, root mean square error and correlation coefficient between the ANFIS model and measured data of flow stress are 0.75%, 2.13 and 0.999 6, respectively, indicating that the ANFIS model can well reflect the real feature of the alloy during practical deforming process. In comparison with the regression model, whose mean percentage error is 6.28% under the same condition, ANFIS parades more accurate prediction performance for flow stress.