Anxiety disorders are among the most common psychiatric diagnoses worldwide and rank prominently among the World Health Organization’s leading causes of disability. Existing treatments for anxiety disorders are inconsistently effective and often cause adverse side effects, underscoring the need to develop clinical entry points that lead to new intervention strategies. In the past decade, clinical and basic research efforts have identified uncertain threat anticipation as causal in anxiogenesis and have established sensitivity to uncertain-threat anticipation as a transdiagnostic marker of anxiety disorders. Across a range of temporally “certain-vs-uncertain” threat paradigms in nonhuman animals and human participants alike, uncertain-threat contexts consistently elicit heightened behavioral, physiological, neurobiological, and (in the case of humans) emotional measures of anxiety. However, the research community has no compelling explanation as to why individuals become more anxious in uncertain-threat contexts—possibly because the term “uncertainty” is sufficiently imprecise as to allow for many operationalizations. This dissertation builds toward a higher-acuity operationalization of “uncertainty” that enables sophisticated manipulations of its tractable features. Chapter 1 details our replication and extension of nonhuman-primate research into the neural substrates of threat processing. We used a novel machine-learning approach to uncover a new link between infant temperament and adolescent behavioral and neurobiological response to uncertain threat (N=18). In Chapter 2, we argue that the extended amygdala has evolved as an arbiter of risk-vs-reward tradeoffs for survival optimization, propose a novel “feature-space” model to explain how distinct pathophysiologies can promote a common clinical phenotype, and discuss the implications of our model in the context of psychiatric disorder. In Chapter 3, we recount our development of a computational model of uncertain-threat anticipation used to decompose “uncertainty” into two tractable features. We held one feature (i.e., discrete threat probability) constant while manipulating the other (i.e., hazard rate) in a statistical threat-learning study (N=42) in which our volunteers made risk-vs-reward tradeoffs in uncertain-threat environments. Through this novel approach, we learned that hazard rate causally drives anxious behavioral and emotional responses during uncertain-threat anticipation, irrespective of discrete threat probabilities. Collectively, our work elucidates the neurocomputational architecture of anxiety.