Abstract: Adaptive decomposition, reconstruction, and denoising of bridge structure monitoring signals are critical parts in the research field of bridge health monitoring. To provide efficient and effective time-frequency domain denoising methods for these signals, an Adaptive Variational Mode Decomposition and Reconstruction (AVMDR) method was proposed for signal denoising, which can overcome the disadvantage of VMD (Variational Mode Decomposition) type methods that the number of decomposition components needs to be determined inadvance. The Empirical Mode Decomposition (EMD) method was introduced to adaptively determine the number of decomposition components, and then the Multi-scale Principal Component Analysis (MSPCA) was used to denoise each component and reconstruct the signal. The denoising performance of the proposed AVMDR method was validated and compared using both simulated signals—linear stationary and nonlinear non-stationary signals with varying noise levels—and real signals obtained from two cable-stayed model bridges. The results indicate that the AVMDR method outperforms other commonly used methods in terms of denoising performance, achieving optimal scores across all denoising performance evaluation metrics. Moreover, the AVMDR method can effectively retain more structural information while eliminating noise.