Most molecular estimates of the age of angiosperms, based on the assumption of a molecular clock, have predated the oldest fossil records of crown-group angiosperms in the Early Cretaceous (135 MA). Previous experiments with rbcL showed marked variation in evolutionary rates among lineages and suggested that ages based on this gene had been overestimated due to reliance on high-rate herbaceous taxa and failure to consider rate variation across sites, and they revealed large, unexplained differences between ages based on different codon positions. Proposed solutions include use of more fossil age constraints on nodes and new methods that allow deviation from a clock. Using all positions of rbcL and assuming a clock, addition of 18 minimum ages within angiosperms and outgroups, which pushes back the ages of some ancient clades (e.g., Proteales) before their inferred molecular ages, increases the estimated age of angiosperms from 141 to 201 MA. Assuming a maximum age of 125 MA for eudicots lowers this only slightly, to 184 MA. Nonparametric rate smoothing (NPRS), which assumes that rates vary gradually, results in ages that conflict more with the fossil record, not less: 207 MA with no constraints, 277 MA with minimum constraints, and 215 MA with maximum and minimum constraints. A new penalized likelihood method designed to overcome possible shortcomings of NPRS gives ages similar to those based on NPRS. These results may mean either that angiosperms are much older than their known fossil record, or that rates of molecular evolution did not change smoothly but rather accelerated during the radiation of angiosperms and slowed in "living fossils" such as Platanus, Nelumbo, and Winteraceae. Plastid photosystem genes show less conflict between dates based on different codon partitions, suggesting that rbcL may not be a good choice for molecular age estimates.

Key words: angiosperms, fossil record, molecular clock, rate smoothing, rbcL