Robust predictions of ecosystem responses to climate change are challenging. To achieve such predictions, ecology has extensively relied on the assumption that community states and dynamics are at equilibrium with climate. However, empirical evidence from Quaternary and contemporary data suggest that species communities rarely follow equilibrium dynamics with climate change. This discrepancy between the conceptual foundation of many predictive models and observed community dynamics casts doubts on our ability to successfully predict future community states. Here we used community response diagrams (CRDs) to empirically investigate the occurrence of different classes of disequilibrium responses in plant communities during the Late Quaternary, and bird communities during modern climate warming in North America. We documented a large variability in types of responses including alternate states, suggesting that equilibrium dynamics are not the most common type of response to climate change. Bird responses appeared less predictable to modern climate warming than plant responses to Late Quaternary climate warming. Furthermore, we showed that baseline climate gradients were a strong predictor of disequilibrium states, while ecological factors such as species’ traits had a substantial, but inconsistent effect on the deviation from equilibrium. We conclude that (1) complex temporal community dynamics including stochastic responses, lags, and alternate states are common; (2) assuming equilibrium dynamics to predict biodiversity responses to future climate changes may lead to unsuccessful predictions.