Geriatric intermediate care facility, GICF, have been established since 1986 in Japan. The term of 'intermediate' denotes a facility midpoint between a hospital and home, with a powerful rehabilitation function. The annual governmental survey on GICF has been applied to all GICFs in Japan since 1989. This survey is a cross-sectional-type survey. For evaluating the role of GICF, we considered a type of cohort study using these surveys for four years by means of a record linkage method. This method is possible, because subjects of a survey for any year become subjects in the survey of the following year, if they have been in a GICF for more than one year. In this paper, for the purpose of preparing for a study of the role of GICF, we report some results about a record linkage for these surveys, and two continuing rates were calculated for these cohorts. Sources were personal records in 1989 to 1992 gathered by The National Survey on GICF in Japan. The record linkage method is as follows. We divided the annual personal records into several files by both the periods of the last admission and the survey year. Three cohorts, i.e. a cohort-89, a cohort-90 and a cohort-91 for persons for whom the first survey was 1989, 1990 and 1991, respectively are thus obtained. Next, we identified individuals from the annual files in each cohort. As personal names were not available, we identified individuals by sex, birth date, date of admission and facility number by the use of a personal computer and a FORTRAN program. Results were as follows. 1. Multiple matched persons were few (0.5% or less), and identification rates were 90% or more, despite the fact that basic linking variables were used. 2. Continuing rates were about 40% and decreased with new cohorts, while those of a fixed cohort were increased as length of stay increased. These reveal that recent admissions who stay continuously at a GICF were decreasing, but the longer the period for which the elderly stays at a GICF, the less possibility that he will be discharged. 3. Between record agreement rates for individuals over two years for unchangeable characteristics such as a place before admission which are expected to be 100% were about 80%. 4. A record linkage method appears to be useful for calculating continuing rates classified by any attribute.