Empirical tests in the 1990s found little evidence of poor countries catching up with rich – unconditional convergence – since the 1960s, and divergence over longer periods. This stylized fact spurred several developments in growth theory, including AK models, poverty trap models, and the concept of convergence conditional on determinants of steady-state income. We revisit these findings, using the subsequent 25 years as an out-of-sample test, and document a trend towards unconditional convergence since 1990 and convergence since 2000, driven by both faster catch-up growth and slower growth of the frontier. During the same period, many of the correlates of growth – human capital, policies, institutions, and culture – also converged substantially and moved in the direction associated with higher income. Were these changes related? Using the omitted variable bias formula, we decompose the gap between unconditional and conditional convergence as the product of two cross-sectional slopes. First, correlate-income slopes, which remained largely stable since 1990. Second, growth-correlate slopes controlling for income – the coefficients of growth regressions – which remained stable for fundamentals of the Solow model (investment rate, population growth, and human capital) but which flattened substantially for other correlates, leading unconditional convergence to converge towards conditional convergence.