ABSTRACT: AIM:In DSM-5, the definitions of substance use disorders (SUD) were changed considerably, yet little is known about the reliability of DSM-5 SUD and its new features. METHODS:The test-retest reliability of DSM-5 SUD and DSM-IV substance dependence (SD) was evaluated in 565 adult substance users, each interviewed twice by different clinician interviewers using the semi-structured Psychiatric Research Interview for Substance and Mental Disorders, DSM-5 version (PRISM-5). DSM-5 SUD and DSM-IV SD criteria were assessed for past year and lifetime, yielding diagnoses and severity levels for alcohol, tobacco, cannabis, cocaine, heroin, opioids, sedatives, hallucinogen, and stimulant use disorders. Cohen's and intraclass correlation coefficients (ICC) assessed reliability for categorical and graded outcomes, respectively. Factors potentially influencing reliability were explored, including inpatient vs. community participant, days between interviews gender, age, race/ethnicity, and SUD severity. RESULTS:DSM-5 SUD diagnoses had substantial to excellent reliability for most substances (? = 0.63-0.94), and moderate for others (hallucinogens, stimulants, sedatives; ? = 0.50-0.59). For graded outcomes (DSM-5 SUD mild, moderate, severe; criteria count 0-11), reliability was substantial to excellent (ICC = 0.74-0.99). Comparisons of DSM-5 SUD and DSM-IV SD reliability showed few significant differences. Reliability of the DSM-5 craving criterion was excellent for heroin (? = 0.84-0.95) and moderate to substantial for other substances (? = 0.49-0.76). The only factor influencing reliability of SUD was severity, with milder disorders significantly more likely to be discordant between the interviews. CONCLUSION:Reproducibility is crucial to good measurement. In a large sample using rigorous methodology, diagnoses and dimensional measures from clinician-administered interviews for DSM-5 SUD were generally highly reliable.