This paper proposes a denosing method for bridge strain signal based on empirical mode decomposition. When the noise appears in some certain frequency and blends with signal, the existing denoising method such as wavelet threshold filtering method can not work well in this regard. This paper carries on empirical mode decomposition method for actual bridge strain signal and then picks out intrinsic mode functions which contain much noise. For these IMFs we carry on specific threshold filtering process method. After this, we use remaining IMFs, processed IMFs and the residual component to reconstruct the signal. The result shows that, this method can eliminate noise of bridge strain signal under the premise of preserving features of original signal.