Abstract We aimed to evaluate the feasibility of using real-world register data for identifying persons with mild Alzheimer’s disease (AD) and to describe their cognitive performance at the time of diagnosis. Patients diagnosed with AD during 2010–2013 (aged 60–81 years) were identified from the Finnish national health registers and enlarged with a smaller private sector sample (total n = 1,268). Patients with other disorders impacting cognition were excluded. Detailed clinical and cognitive screening data (the Consortium to Establish a Registry for Alzheimer’s Disease neuropsychological battery [CERAD-nb]) were obtained from local health records. Adequate cognitive data were available for 389 patients with mild AD (31%) of the entire AD group. The main reasons for not including patients in analyses of cognitive performance were AD diagnosis at a moderate/severe stage (n = 266, 21%), AD diagnosis given before full register coverage (n = 152, 12%), and missing CERAD-nb data (n = 139, 11%). The cognitive performance of persons with late-onset AD (n = 284), mixed cerebrovascular disease and AD (n = 51), and other AD subtypes (n = 54) was compared with that of a non-demented sample (n = 1980) from the general population. Compared with the other AD groups, patients with late-onset AD performed the worst in word list recognition, while patients with mixed cerebrovascular disease and AD performed the worst in constructional praxis and clock drawing tests. A combination of national registers and local health records can be used to collect data relevant for cognitive screening; today, the process is laborious, but it could be improved in the future with refined search algorithms and electronic data.