Within the fifth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5), some diagnoses are now associated with a severity gradient based on the number of diagnostic criteria satisfied. Reasons for questioning the validity of this approach include the implicit assumptions of equal criterion severity and strict additivity of criteria combinations. To assess the implications of heterogeneity of criterion configurations on severity grading, we examined the association between all observed combinations of DSM-5 alcohol use disorder criteria endorsement, at each level of number of criteria endorsed, and multiple validity measures among 22,177 past-year drinkers from Wave 2 of the National Epidemiological Survey on Alcoholism and Related Conditions (NESARC). Substantial variability of implied severity across criteria combinations was observed at each level of endorsement, with nontrivial overlap in implied severity across criterion counts. Findings suggest severity indices are at best imprecise, and potentially misleading. These problems are likely inherent in traditional polythetic approaches to diagnosis and almost certainly applicable to other disorders. Approaches for improving severity grading are proposed.