OBJECTIVE: To examine the longitudinal diagnostic efficiency of the DSM-IV criteria for borderline personality disorder (BPD). METHODS: At baseline, we used semistructured diagnostic interviews to determine criteria and diagnoses; blinded assessments were performed 24 months later with 550 participants. Diagnostic efficiency indices (specifically, conditional probabilities, total predictive power, and kappa) were calculated for each criterion determined at baseline, with the independent BPD diagnosis at follow-up used as the standard. RESULTS: Longitudinal diagnostic efficiencies for the BPD criteria varied, with the criteria of suicidality or self-injury and unstable relationships demonstrating the most predictive utility. CONCLUSIONS: BPD criteria differ in their predictive utility for the diagnosis of BPD when considered longitudinally. These findings have implications both for clinicians who are considering diagnoses and for researchers concerned with forthcoming revisions of our nosological system.