In this model, four broad brain regions are identified: the sensory/association cortices (SC), the lateral and basal lateral amygdala (FA), the basal medial amygdala (BA), and the ventromedial prefrontal cortex (PFC).

The sensory/association cortices signal the perception of stimuli to FA and the PFC. The FA and PFC each form an expectation that the subject will (FA) and will not (PFC) experience an intrinsically fearful stimulus (IFS). The PFC inhibits activation of the FA. The amount of inhibition is proportional to the PFCs confidence that the subject will not encounter an intrinsically fearful stimulus. The modulated signal is transmitted to the BA which then stimulates other brain regions that induce the physical changes associated with fear.

Both the FA and PFC adapt their expectations based on experience. This model uses two scaled geometric sum probability estimation models (PEM) to represent the behavior of the expectation circuits within the FA and PFC. In reality, the PFC and FA probably estimate the probability that the subject will encounter an IFS based on the ease of recall of positive (instances in which the observed stimulus predicted the IFS) and negative (instances in which the observed stimulus did not predict the IFS) memories involving the observed and expected stimuli. The memories associated with positive instances are probably more easily recalled as the amygdala sends signals to the hippocampus that strengthen episodic memory formation during stressful events. Accordingly, the PEM associated with the PFC has an additional decay term that weakens the negative expectations over time, modeling memory decay. This decay term, allows us to model spontaneous fear recovery.

Experimentation suggests that fear extinction does not, principally, involve forgetting fear associations. Rather, it involves learning new associations that suppress previously learned fear associations. Brain imaging experiments suggest that fear expression and suppression are generated by different brain regions; the amygdala (expression) and the ventromedial prefrontal cortex (suppression). The theory that fear extinction does not involve forgetting fear associations was supported by observations of fear recovery, in which it was observed that subjects recover fear faster than they did during fear induction, which suggests that fear associations persist after fear extinction. To model this, we associate two different sensitivity coefficients to the FA PEM and the PFC PEM. This allows us to express the relative stability of fear associations stored within the FA in comparison to those stored within the PFC.

This model uses a simple linear model to represent fear suppression by the PFC. We define a parameter, ki, that defines the maximum proportion of fear generated by the FA that can be suppressed by the PFC. In reality this property corresponds to the strength of the neural projections from the ventromedial prefrontal cortex to the amygdala. The neurons within this pathway are serotonergic, chronic deficiency in serotonin may inhibit the structural development of this pathway contributing to anxiety regulation disorders. Medications such as serotonin reuptake inhibitors (SSRIs) can increase the amount of serotonin available within this region and over time may foster development. Accordingly, we allow ki to evolve over time within the model.

Significantly, our model posits the existence of intrinsically fearful stimuli. It assumes that certain stimuli, such as extreme pain, and fear, are innately anxiogenic. Expectation models, posit that the fear induced by most stimuli however are the result of learned associations. Presented with a non-intrinsically fearful stimlus (NIFS), the amygdala estimates the probability that the NIFS signals an IFS. The fear elicited by the NIFS is proportional to the estimated probability of encountering the IFS.