Despite the increasing evidence of drastic and profound changes in many ecosystems, often referred to as regime shifts, we have little ability to understand the processes that provide insurance against such change (resilience). Modelling studies have suggested that increased variance may foreshadow a regime shift, but this requires long-term data and knowledge of the functional links between key processes. Field-based research and ground-truthing is an essential part of the heuristic that marries theoretical and empirical research, but experimental studies of resilience are lagging behind theory, management and policy requirements. Empirically, ecological resilience must be understood in terms of community dynamics and the potential for small shifts in environmental forcing to break the feedbacks that support resilience. Here, we integrate recent theory and empirical data to identify ways we might define and understand potential thresholds in the resilience of nature, and thus the potential for regime shifts, by focusing on the roles of strong and weak interactions, linkages in meta-communities, and positive feedbacks between these and environmental drivers. The challenge to theoretical and field ecologists is to make the shift from hindsight to a more predictive science that is able to assist in the implementation of ecosystem-based management.