Conceptual restoration models depict strong correlations between structure and function, with both decreasing as an ecosystem is degraded and increasing during restoration. We evaluated the “linear” and “asymptotic” models by measuring diversity and annual net primary productivity (NPP) within four states of a southern Wisconsin floodplain: a remnant (unplowed) wet prairie, two degraded sites (soybean field and invaded prairie), and a restored prairie. Neither model fit our data for aboveground (ANPP), belowground (BNPP), or total (TNPP) productivity. ANPP declined as species richness increased (r = 0.998, df = 2), with highest values for soybeans (1,024 g/m2; two species in 30 0.25-m2 plots) and invaded prairie (937 g/m2; nine species, 99% cover of Phalaris arundinacea), intermediate for restored prairie (712 g/m2; 28 species), and lowest for diverse remnant prairie (571 g/m2; 36 species). In contrast, BNPP was lowest for soybeans (225 g/m2) and highest for remnant prairie (571 g/m2). TNPP in restored prairie (990 g/m2) matched that of the remnant (1,147 g/m2) within 7 years, but root:shoot NPP ratios were quite different (0.39 and 0.99, respectively). Overall, results suggest that the relationship between species diversity and productivity can differ with the component measured (ANPP, BNPP, or TNPP) and that diversity does not ensure high productivity. Because measuring ANPP does not fully test ecosystem-function theory, we recommend assessing BNPP and additional ecosystem processes in future attempts to determine whether adding species will restore more function to degraded ecosystems.