Carcinoembryonic antigen (CEA) is considered to be an embryonic antigen that is reexpressed in carcinomas. However, at the molecular level little is known about fetal forms of CEA. We have studied fetal liver, which was originally considered to contain CEA. A first-trimester cDNA library from fetal liver was screened with CEA-specific probes, and a dominant cDNA clone was identified and sequenced. This 1.7-kilobase cDNA codes for a complete protein of 426 amino acids, of which 34 constitute a leader peptide. Structurally, it can be divided into four immunoglobulin-like domains homologous to CEA (N-A1-A2-B2) and a hydrophobic tail (12 residues). The A and B domains each contain two cysteines; the N domain has none. The protein has seven potential sites for asparagine-linked glycosylation. It is a form of pregnancy-specific beta 1-glycoprotein (PS beta G) but differs from other PS beta G species at the C terminus. The N and A1 domains show 45% and 51% amino acid sequence identity with the corresponding domains of the three CEA family members whose sequences have been determined. Expression studies showed that the cDNA codes for a 72-kDa glycoprotein that reacts immunologically with antisera to CEA, biliary glycoprotein I, and PS beta G. The 72-kDa glycoprotein was released from the transfected cells. At least six mRNA species were identified in human tissues by using this cDNA as a probe. Genomic DNA analysis with an N-domain-specific probe indicated that the number of genes is relatively small.