The aim of this article is to review studies on human anxiety using the startle reflex methodology and to apply the literature on context conditioning in rats to interpret the results. A distinction is made between cued fear (as in specific phobia), a phasic response to an explicit threat cue, and anxiety, a more sustained and future-oriented response not linked to a specific discrete cue. Experimentally, contextual fear, as opposed to cued fear, may best reflect the feeling of aversive expectation about potential future dangers that characterizes anxiety. Following a brief description of the neurobiology of cued fear and context conditioning, evidence is presented showing that anxious patients are overly sensitive to threatening contexts. It is then argued that the degree to which contextual fear is prompted by threat depends on whether the danger is predictable or unpredictable. Consistent with animal data, unpredictable shocks in humans result in greater context conditioning compared to predictable shocks. Because conditioning promotes predictability, it is proposed to use conditioning procedures to study the development of appropriate and inappropriate aversive expectations. Cued fear learning is seen as an adaptive process by which undifferentiated fear becomes cue-specific. Deficits in cued fear learning lead to the development of nonadaptive aversive expectancies and an attentional bias toward generalized threat. Lacking a cue for threat, the organism cannot identify periods of danger and safety and remains in a chronic state of anxiety. Factors that may affect conditioning are discussed.