Ecosystem-level nutrient dynamics during decomposition are often estimated from litter monocultures. If species effects are additive, we can statistically predict nutrient dynamics in multispecies systems from monoculture work, and potential consequences of species loss. However, if species effects are dependent on interactions with other litter species (that is, non-additive), predictions based on monoculture data will likely be inaccurate. We conducted a 3-year, full-factorial, mixed-litter decomposition study of four dominant tree species in a temperate forest and measured nitrogen and phosphorus dynamics to explore whether nutrient dynamics in mixtures were additive or non-additive. Following common approaches, we used litterfall data to predict nutrient dynamics at the ecosystem-level. In mixtures, we observed non-additive effects of litter mixing on nutrient dynamics: the presence of nutrient-rich species in mixture facilitated nutrient release, whereas nutrient-poor species facilitated nutrient retention. Fewer nutrients were released from mixtures containing high-quality litter, and more immobilized from mixtures containing low-quality litter, than predicted from monocultures, creating a difference in overall nutrient release between predicted and actual dynamics in litter mixtures. Nutrient release at the ecosystem-level was greatly overestimated when based on monocultures because the effect of species interactions on nutrient immobilization was not accounted for. Our data illustrate that the identity of species in mixtures is key to their role in non-additive interactions, with repercussions for mineral nutrient availability and storage. These results suggest that predictions of ecosystem-level nutrient dynamics using litter monoculture data likely do not accurately represent actual dynamics because the effects of litter species interactions are not incorporated.