We analyzed the onset of agoraphobia in a large prospective study of the general population. The annual incidence of Diagnostic Interview Schedule/DSM-III agoraphobia is estimated at 22 per 1000 population per year. Latent class analysis of new cases in the at-risk population reveals the existence of two subtypes of agoraphobia. The relationship of the incidence of the two subtypes to sociodemographic and psychopathologic risk factors suggests different origins. The data also show that in two thirds of the 260 new cases of Diagnostic Interview Schedule/DSM-III agoraphobia, onset occurs without a history of panic attack. We analyzed the relationship of these epidemiologic results to earlier clinical findings in terms of possible measurement and sampling differences.