Realistic global changes are very difficult to accurately represent with experimental manipulations. Yet, ecosystem responses to global changes are very difficult to predict without observations from experimental manipulations. Reconciling these constraints is unlikely to be possible in all locations and for all ecosystems, but a multi-pronged approach involving four major components can take advantage of the kinds of observations that are technically feasible. The four components are as follows. First, it is critical to take advantage of model ecosystems, deploying large-scale, multi-factor global change experiments in a range of ecosystems that are empirically tractable, span key gradients of climate and life-form, and provide unusual access to key mechanisms of the ecosystem response.  Second, the multi-factor global change experiments need to be integrated with a substantial number of simpler experiments and observations designed to expand the range of ecosystems and environmental conditions susceptible to analysis based on the complex experiments.  Third, it is essential to interpret the results of both kinds of experiments in the context of trophic interactions and disturbances that may operate at scales larger than the experimental unit. Fourth, we need to integrate the results of components one through three with a modeling framework that handles both well-understood and poorly-understood aspects of the responses. Progress on all four components has been substantial in recent years, but our ability to predict the future status of ecosystems still falls dramatically short of society’s need for information. Key issues for additional focus should include wildfire, invasives, herbivore preferences, and fragmentation. It is also critical that we develop an understanding of the potential for human actions to facilitate ecosystem responses that preserve critical ecosystem goods and services.