BACKGROUND One of the most salient features of Latin American marriages over the last few decades is the stable timing of their union formation, despite educational expansion, the postponement of and retreat from marriage, and the increase in non-marital cohabitation. OBJECTIVE We examine why educational expansion did not influence the aggregated indicators of women's timing of union formation. METHODS We used recently harmonised international census microdata for eight Latin American countries from the 1970s to the 2000s. RESULTS The results from a logistic regression analysis show that this apparent stability was produced by contrasting shifts that occurred in various educational groups. In most countries the postponement effect that was expected from educational expansion was offset by earlier union formation (mostly through non-marital cohabitation) among the least educated (and formally largest) groups, whereas highly educated women showed no change. © 2013 Esteve, López-Ruiz & Spijker.