Lunar calibration is an effective method for visible and near-infrared spectrum remote sensors. Original lunar observation data should be pre-processed into lunar full disk irradiance in order to compare them with a lunar radiation model. This paper describes the pre-processing of lunar data of the MERSI on the FY-3D satellite. The data pre-processing includes two steps. First, the lunar observation data are identified from the massive space view data. Then, the identified lunar original digital count is converted into full disk irradiance. Based on the MERSI observation mode, the lunar zenith and azimuth in the instrument reference coordinate system are used to construct a threshold model. When the zenith and azimuth meet the threshold conditions, the data are identified as lunar image. According to MERSIs imaging geometry and calibration formula, the lunar full disk irradiance can be calculated from two different lunar images. Irradiance from a single-detector multi-scan image needs to be corrected for the over-sample factor, and irradiance from a multi-detector single-scan image needs to be corrected for the radiation response difference between detectors. Based on this method, about 1 minute of lunar observation data among 30 days can be found and identified. The results show that the average difference of irradiance between the two methods is about 0.9%. The pre-processing method presented in this paper can find the original lunar observation data and convert them to full disk irradiance value, which provides a basis for further absolute radiation calibration and error analysis. It can also provide a reference for lunar calibration of other similar remote sensors.