Recent advances in molecular phylogenetics and a series of important palaeobotanical discoveries have revolutionized our understanding of angiosperm diversification. Yet, the origin and early evolution of their most characteristic feature, the flower, remains poorly understood. In particular, the structure of the ancestral flower of all living angiosperms is still uncertain. Here we report model-based reconstructions for ancestral flowers at the deepest nodes in the phylogeny of angiosperms, using the largest data set of floral traits ever assembled. We reconstruct the ancestral angiosperm flower as bisexual and radially symmetric, with more than two whorls of three separate perianth organs each (undifferentiated tepals), more than two whorls of three separate stamens each, and more than five spirally arranged separate carpels. Although uncertainty remains for some of the characters, our reconstruction allows us to propose a new plausible scenario for the early diversification of flowers, leading to new testable hypotheses for future research on angiosperms.

Perianth bilateral symmetry (zygomorphy) has evolved repeatedly from radial symmetry (actinomorphy) throughout angiosperms. Zygomorphy has previously been linked with plant‐pollinator specialization and higher species diversification. However, the exact number of transitions to and from zygomorphy has remained so far unknown. We recorded perianth symmetry from 761 species, selected to represent all 61 orders and 426 currently accepted families of angiosperms and to include all presumed origins of perianth zygomorphy. We then reconstructed the evolution of perianth symmetry on a consensus backbone tree, using parsimony. We found perianth zygomorphy in 32 orders and 110 families. There was a minimum of 130 origins, almost double of what was previously estimated, and 69 reversals to actinomorphy. Among the origins, 2 were in magnoliids, 29 in monocots, 17 in basal eudicots, 35 in superrosids and 47 in superasterids. Among the reversals, 8 were in monocots, 4 in basal eudicots, 18 in superrosids and 39 in superasterids. This study shows that there has been many more origins of perianth zygomorphy and reversals to actinomorphy than previously shown. We then use this new framework to review the developmental evidence of changes in floral symmetry, showing convergence in the early stages of zygomorphy across angiosperms at the developmental level. We also review the evidence on the genetic control of floral symmetry, suggesting that a restricted number of genes has been recruited multiple times independently to achieve zygomorphy. In contrast to its relative homogeneity at the early developmental and molecular level, zygomorphy appears to be highly variable in its morphological expression at anthesis, involving various processes such as perianth part displacement and differentiation. We then review recent hypotheses on the relationship between floral symmetry, floral orientation, and pollination mode in terms of selective advantages and constraints. Our comprehensive angiosperm‐wide reconstruction of floral symmetry evolution provides a new context for future studies on the developmental, functional, and macroevolutionary aspects of floral symmetry.