With more than half of the world’s population living in towns and cities, urban areas get more and more into the focus of humanitarian relief organisations such as ICRC, Médecins sans Frontières (MSF), or SOS Children’s Villages. A key information required for almost any intervention is an estimation of the population numbers for the towns and cities where these organisations operate in. As census data are usually not available or outdated, population numbers have to be estimated by alternative methods such as remote sensing. To do that built-up densities are estimated from high-resolution image data and population numbers are disaggregated proportional to the densities in a top-down approach. Alternatively, population counts per density unit can be aggregated following a bottom-up approach. Both approaches were tested applying normalised Digital Surface Models (nDSM) derived from tri-stereo Pléiade images for Salzburg, Austria and Port-au-Prince, Haiti; the former for testing the quality and stability of the approach in a well-known setting, the latter for testing the approach in a critical environment. Key findings are that satellite-derived nDSMs provide sufficient accuracy for estimating population distributions, as long as reliable information is available for the separation of residential and non-residential urban areas.