The nighttime stable light (NSL) images on board the Operational Line-scan System (OLS) of the Defense Meteorological Satellite Program (DMSP) are useful for extracting large-scale built-up urban areas. However, most NSL-based studies are presently empirical threshold-based approaches. They always overestimate the areas of built-up land in urban regions because of the ‘blooming’ effect of NSL; and overlook small patches in developing towns where the NSL is much lower. In this study, a neighborhood statistics analysis (NSA) method is developed on the basis of the relative spatial variations between neighboring built-up and non-built-up pixels in DMSP-OLS images. It is applied to extract the built-up areas of eight cities in the Pearl River Delta in 1996, 2000, 2005, and 2009. The validations indicate that the total areas of the NSA-mapped results are highly correlated with those from Landsat TM/ETM+ data (R2 = 0.94; p < 0.001). The comparison results, which are evaluated by landscape indices (the landscape shape index (LSI), the contiguity index (CONTIG), and the perimeter area ratio (PARA)), also show good correlations (R2 > 0.46; p < 0.001). In addition, the total NSL of the built-up urban areas is significantly correlated with the statistical population data (R2 = 0.62; p < 0.001), which indirectly confirms the effectiveness of our proposed method.