In order to overcome the defects of traditional wavelet threshold method and improve the quality of bridge health monitoring signals, an improved wavelet threshold de-noising method based on the bridge deflection signals is proposed. First, the improved threshold can adaptively select different thresholds in different decomposition layers. Then, the wavelet coefficients of the bridge deflection signals are modeled to determine the statistical properties of deflection signals. Next, an improved wavelet shrinkage function is obtained based on these statistical properties. The bridge deflection signal de-noising experiments show that the actual de-noising effect is significant, and this proposed de-noising method is suitable for processing the deflection signal of the bridge monitoring system.