The goal of this study is to develop a comprehensive indicator of integration among countries within a supranational system. Integration is not analysed in terms of the growth in interactions among countries, but in terms of the matrix distribution of interactions among countries. Integration can then be indicated in terms of interaction biases among countries measured by the difference between the observed matrix distribution and the hypothetical random distribution. The indicator is applied to data on research collaborations among European research institutions (1993-2000). Evidence is found that the European science system has indeed become more integrated. The higher level of integration has resulted exclusively from a more evenly distributed pattern of European collaborations, while the strong bias towards intra-national collaborations persisted. The results point to the persistence of national science systems. A future research agenda and science policy implications are discussed.