The evolution of automated and semi-automated systems is rendering continuous regulation relatively obsolete, leaving periodic “management” interventions as the main way in which operators exercise control. Consequently, the human is now more frequently required to respond in uncertain, unusual, or “emergency” conditions. Such circumstances connote high stress environments. Consequently, the research reported here investigates expertise at decision making under stress. The source of stress is ubiquitous in occurrence, namely time pressure. We present a process model that explains and predicts the decision behavior of skilled operators as they manage risk under time stress. The model identifies three components of decision making, (1) attention, (2) assessment, and (3) intervention. Attention (1) scans widely among information displays and focuses action narrowly upon one of three procedures for (2) assessing the attended information. Separate procedures assess (α) the risks posed by the environment, (β) risks generated by interacting with the environment, and (α) uncertainty about those risks. The uniquely appropriate intervention (3) is selected by a small set of rules that match heuristically the assessments of risk and uncertainty to a short list of alternative actions. The model is validated with respect to the operation of skilled operators in the domain of currency exchange. In comparing performance versus simulation data, the model identifies the one procedure that resists automation - the assessment of risks posed by the environment. This assessment involves causal arguments that often rely upon extensive domain knowledge. In contrast, attention to displays, heuristic matching, and the procedures for assessing uncertainty and the risk of interaction can be delegated to an automated decision support system. This result has clear implications for the the design of systems to support skilled decision making under emergency conditions: decision support systems for dynamic environments like currency trading must notify the operator of the occurrence of system parameters that require assessments of environmental risk and incorporate these assessments into automated procedures that recommend appropriate interventions.