Recent research has attempted to improve the identification of individuals at-risk of developing schizophrenia to permit targeted early prevention. Two sets of criteria, one characterized by a subgroup of 'basic symptoms' [Klosterkötter, Hellmich, Steinmeyer, Schultze-Lutter (2001) Arch Gen Psychiat 58:158-164] and one by the ultra high-risk model [Miller, McGlashan, Woods, Stein, Driesen, Corcoran, Hoffman, Davidson (1999) Psychiatr Q 70:273-287; Yung, McGorry, McFarlane, Jackson, Patton, Rakkar (1996) Schizophr Bull 22:283-303], have been associated with positive predictive values for later schizophrenia. This paper is a critical discussion of these predictive values. In the first part, the paper demonstrates that the predictive values of at-risk criteria are mediated by a strong enrichment effect and depend considerably on the structure of early detection systems. Further, it shows that these predictive values do not apply to the general population level, where subclinical psychosis shows high prevalence and incidence rates, and that these values may be less predictive in adolescents. In the second part, the paper discusses the need for specific sensitization on several levels of an early detection system and proposes a selected overview of prototypical models already applied in this field.