The main objective of this study was to develop a hybrid neuro-fuzzy system for estimating the magnitude of EMG responses of 10 trunk muscles based on two lifting task variables (trunk velocity and trunk moment) as model inputs. The input and output variables were represented using the fuzzy membership functions. The initial fuzzy rules were generated by the neural network using true EMG data. Two different laboratory-derived EMG data sets were used for model development and validation, respectively. The mean absolute error (MAE) between the actual and model-estimated normalized EMG values was calculated. Across all muscles, the average value of MAE was 8.43% (SD=2.87%) of the normalized EMG data. The larger absolute errors occurred in the left side of the trunk, which exhibited higher levels of muscular activity. Overall, the developed model was capable of estimating the normalized EMG values with average value of the mean absolute differences of 6.4%. It was hypothesized that model performance could be improved by increasing the number of inputs, including additional task variables as well as the subjects' characteristics.