Most national and international agencies producing population projections avoid addressing explicitly the issue of uncertainty. Typically, they provide either a single projection or a set of low, medium and high variants, and only very rarely do they give these projections a probabilistic interpretation. Probabilistic population projections have been developed for specific industrialized countries, mostly the United States, and are based largely on time-series analysis. On a global level, time-series analysis is not applicable because there is a lack of appropriate data, and for conceptual reasons such as the structural discontinuity caused by the demographic transition. Here we report on a new probabilistic approach that makes use of expert opinion on trends in fertility, mortality and migration, and on the 90 per cent uncertainty range of those trends in different parts of the world. We have used simulation techniques to derive probability distributions of population sizes and age structures for 13 regions of the world up to the year 2100. Among other things, we find that there is a probability of two-thirds that the world's population will not double in the twenty-first century.