In November 1988 a random sample of 435 men and women aged 70-75 years from the general population in Roskilde, a provincial town in Denmark, were invited to participate in a study of nutrition and health. Forty-six per cent of the total sample agreed to participate. The total sample could be characterized by socio-economic variables and data on previous hospitalizations obtained from public registers. A subgroup of the non-participants could be further characterized by some of the topics under study using information obtained by telephone interviews. It was found that the non-participants differed from participants by selected health variables. More non-participants than participants had been hospitalized one year before contact. Telephone interviews with non-participants revealed generally poorer self-judged health and less-frequent eating of cooked meals compared with participants. The degree of bias introduced by this selectivity is estimated by weighting and by a minimum/maximum calculation. Review of participation in previous studies of nutrition and health in the elderly shows that problems with non-participation have been treated in various ways. It is concluded that consideration of factors discriminating between participants and non-participants is important for proper estimation of population parameters.