Feeling, or the subjective emotional experience, is a fundamental element of the emotional reaction, yet past attempts to understand the mechanisms of feeling generation remain limited. The current study presents a signal detection theory (SDT) conceptualization of feeling generation. Accordingly, feeling, like other sensations, reflects an outcome of an inner decision regarding the emotional evidence, and, therefore, can be evaluated via 2 processes: evidence differentiation (d′)—the ability to emotionally differentiate between external stimuli, given the essentially noisy evidence—and criterion (c)—the report threshold, or amount of evidence needed to have an intense reportable feeling. According to the model, feelings can be disproportionally intense (false alarms; e.g., emotional overreaction) or disproportionally weak (misses; e.g., failing to detect danger), with the criterion controlling the relative proportion of these “errors.” Results from a novel task indicate that our conceptualization provides a suitable model for valence (pleasant–unpleasant) feeling generation, as reflected in superior model fit relative to plausible alternative models, nonsignificant lack of fit, and by successful experimental tests of a novel prediction regarding contextual influences and related uncertainty. Additional evidence for the external validity of the model shows that SDT parameters, especially the criterion, were meaningfully correlated with relevant emotion regulation and affective style constructs. Implications for the understanding of feeling generation, in general, and in psychopathology, in particular, are discussed.