Although it is widely thought to influence ecosystem processes, there is little consensus on an appropriate measure of functional diversity. The two major perspectives, to date, are to assume that every species is functionally unique, or to assume that some species are functionally identical, such that functional groups exist. Using a continuous measure of functional diversity (FD) derived from the quantitative functional traits of species, we show that the loss of functional diversity from six natural assemblages was rapid compared with rates of loss from comparable simulated assemblages. Loss of FD occurred faster than loss of functional–group diversity in four of the six natural assemblages. Patterns of functional–group diversity loss depended on the number of functional groups and the number of species in an assemblage. Extinctions that occurred first for species with particular traits (e.g. low leaf nitrogen concentration, deep roots and large body size) caused greater loss of FD than expected by chance in four of the six natural assemblages. In two real assemblages, these trait–dependent extinctions had more severe effects on FD than our simulated worst–case extinction scenario. These data suggest that conserving a large proportion of the functional traits of species requires conserving a large proportion of all species.