The rotavirus gene segment coding for the major outer capsid glycoprotein vp7 was amplified directly from stool specimens by the polymerase chain reaction (PCR). Double-stranded RNA extracted from stool samples was used as the template for reverse transcription, which was followed immediately and in the same reaction mix with amplification, using the Taq polymerase. Various conditions were examined to optimize the yield of the amplified gene. The concentrations of MgCl2, dimethyl sulfoxide, and template RNA were critical. The choice of primer pairs allowed amplification of the entire segment or specific portions. By using type-specific primers derived from distinct regions on the gene, we devised a PCR typing method in which each human serotype virus produced a characteristic segment size, readily identifiable in agarose gels. The PCR typing method was applied to 10 rotavirus reference strains, including all 6 known human serotypes (serotypes 1, 2, 3, 4, 8, and 9), and to 34 stool specimens previously serotyped by an enzyme immunoassay with monoclonal antibodies. An absolute correlation was found between the molecular and serologic methods. In addition, 14 stool specimens nonserotypable by an enzyme immunoassay with monoclonal antibodies could be typed by the PCR method. Besides the application for rotavirus detection and typing directly from stools, the PCR method provides a rapid and efficient means of obtaining large quantities of cDNA suitable for sequencing, cloning, and other genetic studies, precluding the need for cell culture and virus purification.