By using of long-term monitoring data of Runyang Suspension Bridge,the improved back-propagation neural networks (BPNNs) are formulated for modeling the correlations between modal frequencies and environmental conditions including wind,temperature and vehicle load.Then,with the correlation models the environmental effects on modal frequencies are quantified and the abnormal changes of measured frequencies are detected by means of the hypothesis tests.Analysis results reveal that BPNN-based correlation models improved by both early stopping and Bayesian regularization techniques exhibit excellent generalization capability.And the developed correlation models can effectively reduce the environmental variability in modal frequencies.The t-test method provides a good capability to detect the damage-induced 0.16% and 0.12% abnormal changes of the 5th and 6th modal frequencies,respectively.Hence,the proposed method is suitable for real-time monitoring of suspension bridge conditions. By using of long-term monitoring data of Runyang Suspension Bridge, the improved back-propagation neural networks (BPNNs) are formulated for modeling the correlations between modal frequencies and environmental conditions including wind, temperature and vehicle load. Chen, with the correlation models the environmental effects on modal frequencies are quantified and the abnormal changes of measured signals are detected by means of the hypothesis tests. Analysis results reveal that BPNN-based correlation models improved by both early stopping and Bayesian regularization techniques exhibit excellent generalization capability. And the so- models can effectively reduce the environmental variability in modal frequencies. The t-test method provides a good capability to detect the damage-induced 0.16% and 0.12% abnormal changes of the 5th and 6th modal frequencies, respectively .ence, the proposed method is suitable for real-time monitoring of suspension bridge conditions.