Most current scientific policies incorporate debates on cities and the geographic organisation of scientific activity. Research on ‘world cities’ develops the idea that interconnected agglomerations can better take advantage of international competition. Thus, the increasing concentration of activities in these cities at the expense of others could be observed by certain scholars using measures based on scientific publications. Others, however, show that an opposite trend is emerging: the largest cities are undergoing a relative decline in a country’s scientific activities. To go beyond this seeming contradiction, this paper provides a global analysis of all countries with papers in the Web of Science over the period 1987–2007. The author’s addresses were geocoded and grouped into agglomerations. Registering of papers was based on the fractional counting of multi-authored publications, and the results are unambiguous: deconcentration is the dominant trend both globally and within countries, with some exceptions for which explanations are suggested.