The Lidar visibility meter is an effective means to detect atmospheric visibility, but its return signals are weak and easy to be interfered by various noises. In order to improve the accuracy of visibility inversion, the suppressing background noise and extracting useful signals from it are very important. The empirical mode decomposition algorithm is used to decompose and reconstruct the return signals, which achieves good denoising effect. The simulation results show that the empirical mode decomposition algorithm improves the output signal noise rate of return signals effectively and reduces the root mean square error. By processing several groups of measured data under different weather conditions, such as sunny, cloudy, and fog days, the inversion results are compared with the measurement results of atmospheric transmittance LT31, which further verifies the effectiveness of the algorithm.