This article gives an overview of a nationally representive public health research database in Denmark, the Danish National Cohort Study (DANCOS). DANCOS combines baseline data from health interview surveys with both pre- and post-baseline data from national health registries with date from a re-interview survey. As part of the national health interview survey programme, three nation-wide surveys were conducted in 1986/1987, 1991 and 1994. The samples in the three cohorts consisted of 23,096 adult Danes ages 16 years and older, and 18,296 (79.2%) were personally interviewed on socio-economic status; living conditions; health behaviour; self-rated health; morbidity; utilization of health services; and the consequences of illness and disability. Each Dane is equipped with a unique identification number that allows highly accurate linkage of data in the health surveys and in the national health and administrative registries. All respondents and non-respondents were followed through 2002, a total of 3,796 had died and 249 had emigrated. The specific cause of death for 2,485 people was recorded in the Danish Register of Causes of Death, updated through 1998. For 1978-1977, the Danish National Hospital Register contained information about 16,503 persons who had been hospitalized for any illness or related health problems, accounting for 70,925 hospital admissions. For 1970-1997, the Danish Psychiatric Central Register comprised information on 1691 persons. For 1990-1997, the Danish National Health Service Register recorded 980,043 contacts with general practitioners and specialist physicians. For 1943-1996, the Danish cancer Registry contained information about one or more cancer diseases among 1,432 people. A total of 4,334 people in the 1994 cohort were re-interview in 2000. DANCOS allows for a wide range of analysis in a historical prospective design of determinants of morbidity and mortality, of health care utilization and of the social effects of ill health. DANCOS also allows studies of methodological issues, including analyzing the characteristics of non-respondents.