Ecosystem services (ES) are an increasingly popular policy framework for connecting biodiversity with human well-being. These efforts typically assume that biodiversity and ES covary, but the relationship between them remains remarkably unclear. Here we analyse >500 recent papers and show that reported relationships differ among ES, methods of measuring biodiversity and ES, and three different approaches to linking them (spatial correlations, management comparisons and functional experiments). For spatial correlations, biodiversity relates more strongly to measures of ES supply than to resulting human benefits. For management comparisons, biodiversity of 'service providers' predicts ES more often than biodiversity of functionally unrelated taxa, but the opposite is true for spatial correlations. Functional experiments occur at smaller spatial scales than management and spatial studies, which show contrasting responses to scale. Our results illuminate the varying dynamics relating biodiversity to ES, and show the importance of matching management efforts to the most relevant scientific evidence.