Urban villages comprise mainly low-rise and congested, often informal settlements surrounded by new constructions and high-rise buildings whereby structures can be very different between neighboring areas. Monitoring urban villages and analyzing their characteristics are crucial for urban development and sustainability research. In this study, we carried out a combined analysis of multispectral GaoFen-1 (GF-1) and high resolution TerraSAR-X radar (TSX) imagery to extract the urban village information. GF-1 and TSX data are combined with the Gramshmidt spectral sharpening method so as to provide new input data for urban village classification. The Grey-Level Co-occurrence Matrix (GLCM) approach was also applied to four directions to provide another four types (all, 0°, 90°, 45° directions) of TSX-based inputs for urban village detection. We analyzed the urban village mapping performance using the Random Forest approach. The results demonstrate that the best overall accuracy and the best producer accuracy of urban villages reached with the GLCM 90° dataset (82.33%, 68.54% respectively). Adding single polarization TSX data as input information to the optical image GF-1 provided an average product accuracy improvement of around 7% in formal built-up area classification. The SAR and optical fusion imagery also provided an effective means to eliminate some layover, shadow effects, and dominant scattering at building locations and green spaces, improving the producer accuracy by 7% in urban area classification. To sum up, the added value of SAR information is demonstrated by the enhanced results achievable over built-up areas, including formal and informal settlements.