Ecosystems may occupy functionally distinct alternative states, some of which are more or less desirable from a management standpoint. Transitions from state to state are usually associated with a particular trigger or sequence of triggers, such as the addition or subtraction of a disturbance. Transitions are often not linear, rather it is common to see an abrupt transition come about even though the trigger increases only incrementally; these are examples of threshold behaviors. An ideal monitoring program, such as the National Park Service’s Inventory and Monitoring Program, would quantify triggers, and be able to inform managers when measurements of a trigger are approaching a threshold so that management action can avoid an unwanted state transition. Unfortunately, both triggers and the threshold points at which state transitions occur are generally only partially known. Using case studies, we advance a general procedure to help identify triggers and estimate where threshold dynamics may occur. Our procedure is as follows: (1) Operationally define the ecosystem type being considered; we suggest that the ecological site concept of the Natural Resource Conservation Service is a useful system, (2) Using all available a priori knowledge to develop a state-and-transition model (STM), which defines possible ecosystem states, plausible transitions among them and likely triggers, (3) Validate the STM by verifying the existence of its states to the greatest degree possible, (4) Use the STM model to identify transitions and triggers likely to be detectable by a monitoring program, and estimate to the greatest degree possible the value of a measurable indicator of a trigger at the point that a state transition is imminent (tipping point), and values that may indicate when management intervention should be considered (assessment points). We illustrate two different methods for attaining these goals using a data-rich case study in Canyonlands National Park, and a data-poor case study in Wupatki National Monument. In the data-rich case, STMs are validated and revised, and tipping and assessment points are estimated using statistical analysis of data. In the data-poor case, we develop an iterative expert opinion survey approach to validate the degree of confidence in an STM, revise the model, identify lack of confidence in specific model components, and create reasonable first approximations of tipping and assessment points, which can later be refined when more data are available. Our goal should be to develop the best set of models possible given the level of information available to support decisions, which is often not much. The approach presented here offers a flexible means of achieving this goal, and determining specific research areas in need of study.