OBJECTIVES: To develop appropriate algorithms for identifying and classifying relapses among patients with multiple sclerosis in claims databases.   METHODS: Algorithms for detecting relapses in claims databases were identified by literature searches and refined by neurologists and database experts to identify relapses in inpatient and outpatient settings. Definitions were used in the US Department of Defense (DOD) and MarketScan® databases to determine annualized relapse rates (ARR) occurring after index date (date of second MS medical claim separated by ≥30 days, following ≥12 months of continuous enrollment) during the 12-month and all available follow-up time. RESULTS: A relapse was defined as an inpatient visit with a primary ICD-9-CM diagnosis code 340.xx or both an outpatient visit with any 340.xx diagnosis code and oral or intravenous corticosteroid use ≤7 days of the outpatient visit. ARR estimates in the DOD (N=15,447) and MarketScan® (N=35,134) databases were 0.25–0.30 and 0.20–0.27, respectively. For inpatient relapses, estimates were 0.04–0.06 in the DOD database and 0.02–0.03 in the MarketScan® database. The corresponding estimates for outpatient relapses were 0.20–0.23 and 0.18–0.24. Severe relapses required an inpatient visit plus additional evidence of brain or spinal magnetic resonance imaging ≤7 days before or during hospitalization, or an outpatient visit combined with extended treatment (starting 0–30 days after outpatient visit) with ≥1 of the following: a further course of intravenous corticosteroids >7 days after the first course; a course of oral corticosteroids >7 days after the intravenous course; intravenous immunoglobulins or plasma exchange. Algorithms for identifying atypical relapses were also developed. CONCLUSIONS:  General and outpatient ARRs are consistent between the two distinct claims databases and are similar to those reported in the literature. Differences in inpatient ARRs may indicate differences between clinical practices in the two systems. Further investigation in the real-world setting is required.