In order to predict the loom efficiency more accurately in the weaving workshop of textile mills, three models, i.e. BP neural network, principal component analysis combined with BP neural network(PCA-BP) and genetic algorithm modified BP neural network model (GA-BP), were used to predict the loom efficiency. At the same time, the prediction results of the GA-BP were compared with that of the BP neural network and PCA-BP neural network. The results show that the GA-BP has the best fitting degree to the original data, the correlation coefficient is 0.946 87, which is 6.42% higher than BP and 2.61% higher than PCA-BP. The average absolute errors between the simulated output value and the expected loom stoppage values over 100 000 weft insertions are 0.341 2, 0.303 1 and 0.234 1, respectively, for GA-BP, PCA-BP and BP models, corresponding to error percentages 8.63%, 7.67% and 5.92%. The average errors between the predicted and the expected values of the loom efficiency with different network models are 3.010 9, 2.688 4 and 2.118 9, respectively, with error percentages of 3.51%, 3.13%, 2.47%. The order of prediction accuracy of the three models is GA-BP, PCA-BP and BP.