The Revised Diagnostic Interview for Borderlines (DIB-R; Zanarini, Gunderson, Frankenburg, & Chauncey, 1989) measures four major aspects of borderline personality disorder (BPD): Affect, Cognition, Impulse Action Patterns, and Interpersonal Relationships. In the present study, 353 young adults completed the DIB-R at age 18 (Wave 1) and again two years later (Wave 2) at age 20. Concerning the prediction of future BPD features, three models were compared: (a) Wave 1 Affect scores predicting all Wave 2 BPD features (NA model); (b) Wave 1 Impulse Action Patterns scores predicting all Wave 2 BPD features (IMP model); and (c) both Wave 1 Affect and Impulse Action Patterns scores predicting all Wave 2 BPD features (NA-IMP model). Each model controlled for stabilities over time and within-time covariances. Results indicated that the NA model provided the best fit to the data, and improved model fit over a baseline stabilities model and the other models tested. However, even within the NA model there was some evidence that the impulsivity scores were not accounted for by other BPD features. These results suggest that although negative affect is predictive of most BPD symptoms, it does not fully predict future impulsive behavior.