This paper proposes an easy‐to‐use nonparametric indicator for club convergence, or convergence within clusters of countries: it measures whether the modes of the gross domestic product (GDP) per capita distribution become more pronounced over time. Relying on changes in the critical bandwidth for unimodality, the indicator is a dynamic extension of concepts from often‐used multimodality tests. Its evolution suggests the new empirical result of a ‘millennium peak’ in club convergence in the worldwide GDP per capita distribution. The club convergence movements of the 1980s and 1990s, when groups of poor and rich countries converged to two separate points, was followed by a de‐clubbing movement after the turn of the millennium. Copyright © 2016 John Wiley & Sons, Ltd.