Scanning lidar can produce a profile image about atmosphere and play an important role in studying boundary layer, structure of cloud, distribution and transport of air pollution. However, traditional Fernald and Klett methods cannot be applied to the situation of large zenith angles. Usually, by using two-angle method, the lidar constant is corrected, and then atmospheric extinction coefficient is obtained. But this is a multiple-solution problem. It is difficult to optimize the solutions just by setting some relevant conditions. The atmospheric region between adjacent slant paths is basically horizontally uniform. The lidar constant is estimated by using linear regression in that region. Then the best lidar constant is optimized with relevant constraints. The atmospheric extinction coefficient is inverted. By theoretical simulated and measured data analysis, the inversion result is still very good even when the zenith angle is large and lidar signal quality is not very good. Results show that this method can display atmospheric structure very well.