This article addresses the clinical issue of selecting assessments of attachment that are relevant to decision making for families. The validity of three commonly used methods of assessing attachment in preschool-aged children was compared using a sample of 51 low-income mother–child dyads. Thirty-eight of the children had been abused or neglected. The dyads were seen in a Strange Situation that was classified using each of the three methods: (a) the Ainsworth-extended method, (b) the Cassidy-Marvin (C-M) method, and (c) the Preschool Assessment of Attachment (PAA). Validity was evaluated in terms of maltreatment status, maternal sensitivity, child DQ, and maternal attachment strategy. The PAA and C-M classifications matched in only 37% of cases. The Ainsworth-extended method differentiated secure versus insecure children on two variables. The C-M method differentiated secure versus insecure children on one variable. The PAA differentiated secure versus insecure children on all four variables and subgroups on one. Moreover, it was tied to other family relationship variables in meaningful ways. If applied in clinical settings, these three methods would result in very different groups of children being seen as safe and at risk. We argue that clinicians cannot afford to be uninformed about the validity of alternative means of assessing attachment.