From a theoretical perspective crop yields should tend to converge over time and space as: growth in yield potential exhibits diminishing returns; an increasing share of farmers shift to using high yielding varieties (HYVs); barriers to the free flow of knowledge and information are removed; and significant investments continue to be made in supporting institutions whose mandates include facilitating and accelerating the cross-border flow of improved agricultural technologies. Using a new, sub-national crop yield database for Latin America and the Caribbean (LAC) we examine whether convergence is indeed occurring, and discover it is not. On the contrary, there is evidence of divergence. We test three hypotheses that might help account for this finding: that technology generation has been biased towards production in more-favored production systems leaving behind persistent pockets of production in more marginal lands; that rainfall patterns have changed in ways that exacerbate yield divergence, and that technology spillover across borders remains more problematic than within-country spillover. We find evidence to support all three of these hypotheses. Further work is needed to assess the relative importance of these sources of yield divergence both across and within LAC. As anticipated, rainfall variability is poorly linked to the variability of irrigated crop yields, but more strongly linked to variability in rainfed crops. The results suggest while some countries and regions within countries forge ahead with crop yield improvements there are many areas, often in smaller countries, where the livelihoods of many farmers - and likely a disproportionate share of LAC’s rural poor - continue to be constrained by low-productivity agriculture. There remains significant work ahead for national governments and for publicly-funded regional and international agricultural technology institutions to remedy this situation.