We examined the association between risk of dementia or Alzheimer's disease (AD) and occupation by using measures of complexity of work with data, people, and things. The study included 10,079 members of the population-based Swedish Twin Registry who were participants in the HARMONY study. We diagnosed dementia by means of a two-stage procedure--cognitive impairment screening followed by full clinical evaluation. We analyzed data with case-control and cotwin control designs. The cotwin control design provides control over genetic and familial factors. In the case-control study, controlling for age, gender, and level of education, we found that more complex work with people was associated with reduced risk of AD. Greater complexity of work with people and data was protective in twin pairs discordant for AD. Findings suggest that greater complexity of work, and particularly complex work with people, may reduce the risk of AD.