Abstract
Many controversial questions about the shape and evolution of city size distributions can be solved if reliable, large and comparable set of date are used for several countries. We provide new empirical evidence by using the large data base ‘Geopolis’, which has strictly comparable figures for all towns and cities of the world over 10000 inhabitants between 1950 and 1990. A Pareto model is used for identifying as metropolises one or a few large cities for each national urban system. From those data, two empirical power laws are established, linking the size of the metropolises to the size of their national urban system. The first is a transversal law : for a set of countries at a given date, the share of population concentrated in metropolises tends to decrease when larger countries are considered. The second law, which is longitudinal, shows that metropolises in the past have grown in a systematic way more rapidly than the rest of their urban system, invalidating Gibrat’s urban growth model. Such empirical regularities could help for predicting the future of nowadays observed metropolisation trends.