To examine the effects of the 1989 Medicaid-elibigility expansion for pregnant women in Florida, the authors constructed a linked person-level database that contained information from eight different files. This report describes each of the sources and then details the linking procedure and the resulting analytic database. The authors used two Vital Statistics files (VS) to define the universe of Florida deliveries. The birth certificate file and the fetal death file contained demographic information about the mother and baby as well as prenatal care and birth outcome data. The birth certificate file was linked to the death certificate file to identify babies who died within the first year of life. The authors matched the VS records to the Hospital Discharge file to obtain insurance information. The authors developed a matching algorithm using hospital, patient's date of birth, date of first procedure, and patient's zip code. Approximately 88 percent of the VS records linked to a mother's discharge record according to the match criteria. Using Social Security number, the authors matched the linked file to two additional files to obtain supplemental information for those using the public health system and those who were eligible for Medicaid. Seventy-six percent of women with prenatal care records on the Public Health System Encounter file matched to the linked file. Eighty percent of those identified as Medicaid on the linked file matched to a Social Security number on the Medicaid-Eligibility file. Hospital characteristics were added for the 98 percent of cases the authors matched to the American Hospital Association database. Socioeconomic characteristics of the mother's neighborhood were included for the 95 percent of women with zip codes that matched to the 1990 census. The authors' matching algorithm required exact matches when linkage variables were unique, such as Social Security number. It was more lenient for variables that are coded less reliably, such as zip code. This approach allowed the authors to maintain confidence in the reliability of the data without sacrificing the sample size. The final analysis files were rich with information from many sources and contained an average of 83 percent of the original population of births and fetal deaths.