Abstract Methods were developed to extract population and dwelling unit data from a high-resolution multi-spectral SPOT image using an integrated digital image analysis and GIS approach for the Hong Kong metropolitan area, which is characterized by high population densities and mixed land use located in multifunctional structures. These methods include different forms of regression models linking spectral radiance values of image pixels with population densities, as well as with high and low residential densities. It was found that accurate estimates of population and dwelling units could be obtained at the macro level for the whole study area using any one of these models. However, at the micro level, the accuracy tended to be low because of the difficulty in discriminating residential use from non-residential use in multifunctional buildings in a satellite image.