Despite extensive research efforts, the controversy over diversity–productivity (D–P) patterns in natural communities still looms large. Recent meta-analyses suggest that unimodal D–P relationships tend to pre-dominate in plant studies, while positively linear relationships are more common in animal studies. These patterns, however, are based on studies in which productivity is estimated either directly, based on the biomass or energy of the studied organisms, or indirectly, according to the productivity of lower trophic levels, and various surrogates. Our analysis shows that the distribution of D–P patterns is sensitive to the directness of productivity estimates in animal studies but not in plant studies. Analysis of D–P patterns should be based on direct productivity estimates of the studied organisms, especially in comparative meta-analyses of communities from multiple trophic levels, where productivity is often affected nonlinearly by indirect factors or when complex feedback interactions are expected between productivity and diversity.