The mixing heights determined by the micropulse lidar and radiosonde profile data on sunny days (from December 2016 to November 2017) in Chengdu are compared. The error analysis of mixing heights based on lidar is made. The results show that: The mixing heights determined by micropulse lidar and radiosonde data have good consistency (the correlation coefficient is 0.75), which indicates that it is accurate to use the micropulse lidar data to invert the mixing heights in Chengdu. The mixing heights inverted by lidar data are slightly lower than the values determined by radiosonde data. The deviation of the values which are determined by the two methods is very small (about 20%) when the mixing height is from 1000 m to 2000 m. The deviation increases slightly (26%) when the mixing heights are lower than 1000 m or higher than 2000 m. The variation trend of mixing heights determined by the two methods is relatively consistent, which are lower in December and January, and higher in April and May. The mixing heights determined by lidar have obvious diurnal variation characteristics. The mixing heights increase rapidly in the morning and the growth rates slowdown in the afternoon. The mixing heights reach the maximum in the afternoon, and decrease rapidly after sunset. The increase of relative humidity and the existence of the residual layer are the reasons that lead to the large error in the inversion of the mixing layer height based on lidar data.