OBJECTIVES : It is often difficult to examine dementia subpopulations, limiting the ability to study, treat, or stratify by clinically different etiologies. The objective was to develop an algorithm using administrative claims data to classify individuals with various subtypes of dementia. METHODS : Using de-identified data from OptumLabs^TM Data Warehouse, we defined a cohort of cases with a first, new dementia diagnosis from 2011 to 2014 using ICD-9 codes and prescription claims. To classify cases by dementia subtype, a set of decision rules were developed based on temporal sequencing of diagnoses, anti-dementia prescription fills, and provider specialty six months following index diagnosis with input from a technical expert panel. Dementia subtypes included: Alzheimer’s (AD); mild cognitive impairment (MCI); Lewy-body (LBD); frontotemporal (FTD); vascular (VD); and non-specific (NSD). Individuals with no dementia diagnosis but with anti-dementia prescription fills were assigned to a drug-only category. Two additional categories, multiple diagnoses (Multi) and multiple diagnoses by specialists (Multi-S) were added in response to observed data. Demographic and comorbidity data were used to assess the construct validity of the categorizations. Descriptive analyses were conducted. RESULTS : The algorithm categorized 36,838 cases of dementia into subtypes and diagnostic categories. The majority of cases were NSD (41.2%), followed by drug-only (15.6%), MCI (15.3%), AD (10.2%), Multi (8.5%), VD (4.1%), Multi-S (4.1%), LBD (0.6%), and FTD (0.4%). Categorizations were supported by demographic and comorbidity data (e.g., MCI and FTD with youngest age groups with lowest-percent of Medicare coverage; VD with highest-percent of cardiovascular comorbidities; LBD with highest-percent of Parkinson’s disease). CONCLUSIONS : Dementia cases identified from administrative claims data can be categorized into dementia subtypes using an algorithm, which relies on diagnostic and prescription drug codes, temporal sequencing, and provider specialty. Additional validation of the algorithm will help support understanding how and where it can be applied in health economics and outcomes research.