The inevitable coarse pixels (~25 km) of satellite passive microwave images introduced large uncertainty to the surface melt area estimation on Antarctic ice margins. Our test showed that the melt index of the Austral year 2012-13 in the Antarctic Peninsula calculated from the high resolution product was 33% lower than the original Special Sensor Microwave/Imager (SSM/I) images. Therefore, by allowing for fractional melt estimation, a subpixel mapping method was adopted in this research to improve the accuracy and reliability of surface melt measurement from passive microwave images. This innovative method uses the least squares mixture analysis (LSMA) on the time series of daily passive microwave images by taking advantage of their high temporal resolution. The endmembers for the unmixing calculation were collected under the constraint of voronoi polygons. The fractional melt index of each pixel was calculated by multiplying its area with melt fraction. By using the high resolution passive microwave earth system data record (PMESDR) dataset as the reference, we found that compared with the original SSM/I images, the overestimation of surface melt was corrected by the unmixing analysis. A log-linear regression between melt fraction and elevation showed that the melt fraction is inversely correlated to the elevation, and the topography is the dominant factor for melt fraction distribution in high elevations. We recommend such a treatment of linear unmixing analysis on the passive microwave images to be used for future surface melt mapping in Antarctica and Greenland.