Experimental evidence that plant species diversity has positive effects on biomass production appears to conflict with correlations of species diversity and standing biomass in natural communities. This may be due to the confounding effects of a third variable, resource availability, which has strong control over both diversity and productivity in natural systems and may conceal any positive effects of diversity on productivity. To test this hypothesis, I independently manipulated resource availability (soil fertility) and sown species diversity in a field experiment and measured their individual and interactive effects on productivity. Although fertility was a far stronger predictor of productivity than diversity, the effect of diversity on productivity significantly increased with fertility. Relative yield analyses indicated that plant mixtures of high fertility treatments significantly "overyielded," or were more productive than expected based on monoculture yields of component species. In contrast, plant mixtures of low fertility treatments had significantly lower-than-expected yields. The effect of diversity on productivity was also driven by sampling effects, where more species-rich mixtures were more likely to include particularly productive species. Unexpectedly, the strength of sampling effects was largely insensitive to fertility, although the particular species most responsible for sampling effects did change with fertility. These results suggest that positive effects of species diversity on ecosystem productivity in natural systems are likely to be masked by variation in environmental factors among habitats.