Abstract Introduction The objective was to examine whether subjective memory impairment (SMI) predicts all-cause dementia or Alzheimer's disease (AD) in a population-based study with long-term follow-up (median = 10 years). Methods A total of 2043 initially dementia-free participants (≥ 60 years) made three memory ratings (“compared with others”, “compared with five years ago”, and “complaints from family/friends”) at baseline. During follow-up, 372 participants developed dementia (208 with AD). Results Cox regression revealed that subjective memory impairment ratings predicted all-cause dementia in models adjusting for age and sex (hazard ratio or HR from 2.04 to 3.94), with even higher values for AD (HR from 2.29 to 5.74). The result persisted in models including other covariates, including baseline episodic memory performance, and in analyses restricted to participants with long time to dementia diagnosis (≥ 5 years). Discussion The findings underscore the usefulness of subjective memory assessment in combination with other factors in identifying individuals at risk for developing dementia.