While the relationship between food web complexity and stability has been well documented, how complexity affects productivity remains elusive. In this study, we combine food web theory and a dataset of 149 aquatic food webs to investigate the effect of complexity (i.e., species richness, connectance, and average interaction strength) on ecosystem productivity. We find that more complex ecosystems tend to be more productive, although different facets of complexity have contrasting effects. A higher species richness and/or average interaction strength increases productivity, whereas a higher connectance often decreases it. These patterns hold not only between realized complexity and productivity but also characterize responses of productivity to simulated declines of complexity. Our model also predicts a negative association between productivity and stability along gradients of complexity. Empirical analyses support our predictions on positive complexity-productivity relationships and negative productivity-stability relationships. Our study provides a step forward toward reconciling ecosystem complexity, productivity and stability.