Building density, as a component of impervious surface fraction, is a significant indicator of population distribution as essentially all humans live and conduct activities in buildings. Because population spatialization usually occurs over large areas, large-scale building density estimation through a proper, time-efficient, and relatively precise way is urgently required. Therefore, this study constructed a decision tree by the Classification and Regression Tree (CART) algorithm combining synthetic aperture radar (SAR) with optical images. The input features included four spectral bands (B1–4) of GF-1 PMS imagery; Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), and Ratio Built-up Index (RBI) derived from them; and backscatter intensity (BI) of Radarsat-2 SAR data. In addition, a new index called amended backscatter intensity (ABI), which takes the influence created by different spatial patterns into account, was introduced and calculated through fractal dimension and lacunarity. Result showed that before the integration use of multisource data, a model using B1–4, NDVI, NDWI, and RBI had the highest accuracy, with RMSE of 10.28 and R2 of 0.63 for Jizhou and RMSE of 20.34 and R2 of 0.36 for Beijing. In Comparison, the best model after combining two data sources (i.e., the model employing B1–4, NDVI, NDWI, RBI and ABI) reduced the RMSE to 8.93 and 16.21 raised the R2 to 0.80 and 0.64, respectively. The result indicated that the synergistic use of optical and SAR data has the potential to improve the building density estimation performance and the addition of ABI has a better capacity for improving the model than other input features.