Despite much scrutiny the relationship between productivity and species richness remains controversial, and there is little agreement about causal processes. We present the results of a survey of 159 productivity-plant species richness (P-PSR) relationships from 131 published studies. We critically assessed each study with respect to experimental design and for the appropriateness of the surrogates used for productivity. We were able to accept only 60 of the reported relationships as robust tests of the P-PSR relationship and a further 18 as robust tests of the biomass species richness relationship. Previous analyses have found that unimodal P-PSR relationships predominate. In contrast, we found that, in studies that used data of continental to global extent, all P-PSR relationships were positive regardless of grain, that almost all were also positive in data sets of regional extent, and that unimodal relationships were not dominant even in studies of fine grain or small spatial extent. Our results differ substantially from previous meta-analyses because previous studies have included a large number of studies that do not meet basic experimental design criteria for objectively testing P-PSR relationships. These results have important implications for theory that attempts to explain species richness patterns. We critically review four dominant theories in light of our results and develop new falsifiable predictions of relationship from these theories at both small and large spatial scales.