Objective: The purpose was to identify different operational definitions of maltreatment severity, and then to examine their predictive validity. Method: Children and their primary caregivers participating in a consortium of ongoing longitudinal studies were interviewed when they were approximately 4 and 8 years of age to assess behavior problems, and developmental and psychological functioning. Four different severity definitions were identified and applied to 519 children who were reported for alleged maltreatment between Birth and the Age 8 interview. A taxonomy for defining maltreatment characteristics (Barnett, Manly, & Cicchetti, 1993) was applied to Child Protective Service records to define severity as (a) Maximum Severity within each of five maltreatment types, (b) Overall Maximum Severity across the five types, (c) Total Severity or the sum of the maximum severity for each of five types, and (d) Mean Severity or the average severity for those types of maltreatment alleged, during each of two time periods - Birth to Age 4, and Age 4 to Age 8. Results: Regression analyses that controlled for socio-demographic factors, early maltreatment (Birth to Age 4), prior functioning (Age 4), and site revealed that (a) all four severity definitions for maltreatment reports between Age 4 and Age 8 predicted Age 8 behavior problems, (b) Maximum Severity by Type and Mean Severity predicted adaptive functioning at Age 8, and (c) only Maximum Severity by Type was related to anger, at Age 8. Follow-up regression analyses indicated that only Maximum Severity by Type, specifically physical abuse, accounted for outcomes, beyond maltreatment occurrence versus non-occurrence. Conclusion: The results suggest that maltreatment severity definitions that preserve ratings within types of maltreatment may be the optimal approach to measure the severity of children's experiences. Copyright 2005 Elsevier Ltd. All rights reserved.