High data transmission and storage cost caused by massive data collected is a critical problem in bridge structural health monitoring. Thus the compressed sensing theory is introduced in this paper to reduce the sampling. The conventional measurement matrix is optimized to reduce the coherence between the measurement matrix and sparse basis，which is benefit for accurately reconstructing the original dynamic responses with a limited sampling data. The field test data under ambient vibration of Ji'an Bridge is utilized to verify the feasibility and effectiveness of the proposed structural dynamic response reconstruction method for bridges based on compressed sensing. The studied results include：the reconstructed response based on compressed sensing agrees well with the original response in time domain；when the compression ratio is more than 20%，the relative errors of the reconstructed response maintain less than 10%；the reconstructed responses using optimized measurement matrix provide higher accuracy than those using initial measurement matrix，especially in the case of lower compression ratio，resulting in reduction of data collection；the spectrum of the reconstructed response using optimized measurement matrix is smoother and matches well with the original response spectrum，and the peaks of the spectrum can be picked accurately；in contrast，the spectrum of the reconstructed response without optimization of the measurement matrix has more misjudgment of peaks，and some peaks even cannot be identified；the measurement matrix optimization method can be applied to random Gaussian matrix，Bernoulli matrix and sparse random matrix.The results demonstrate that the proposed structural dynamic response reconstruction method for bridges based on compressed sensing is an effective way to accurately reconstruct original dynamic response with a limited sampling data.