The energy received through remote sensing sensors contains the amount of reflected solar energy and emitted energy of objects in middle-wave infrared (MWIR, 3~5 μm). Usually, the reflected solar energy is weak in MWIR spectrum. In some certain situations like sun glint area in sea surface, however, the energy is relatively significant and less sensitive to atmospheric effects. Meanwhile, for the satellite sensor which equipped with onboard calibration system, its onboard radiation performance of MWIR（using blackbody calibration)is quite stable. Therefore, the MWIR reflectance in sea surface glint area can be considered as a reference for cross-calibration between the solar reflected bands. Based on this idea, this paper constructed an improved non-linear split window model that is suitable for VIIRS (visible infrared imaging radiometer) MWIR band and used this model to calculate the MWIR reflectance of sun glint area in southern Indian Ocean. This model made statistics, getting the relationship between the reflectance of VIIRS M12 and M13 bands at first, and then used the non-linear split window algorithm to calculate the actual sea surface reflectance. The uncertainty of the simulation model was 083%. On this basis, this paper calculated sea surface reflectance of selected sample regions based on the data of VIIRS M12 band (center wavelength: 3697 μm) in sun glint areas. And then verified the reflectance accuracy by two methods, getting the two accuracies were about 0239% and 023%, respectively. It proves that the calculation model in this paper can greatly improve the accuracy compared to the situation when the sea surface reflectance is between M12 and M13 which are assumed to be equal (accuracy of 248% and 103%, respectively). It also indicated that the model is feasible and effective to calculate the reflectance in sea surface glint area with VIIRS M12 MWIR band, and the accuracy can meet the requirements of MWIR sea surface reflectance as a calibration reference among bands.