The polymerase chain reaction (PCR) for human immunodeficiency virus (HIV) was evaluated using coded blood specimens from infants whose clinical status is now known. A micromethod for the efficient isolation of mononuclear cells from small volumes of blood, and definitions of PCR positivity that took into account the number and purity of these mononuclear cells, were established in an attempt to define parameters for quality assurance. Results of HIV culture, p-24 antigen, and HIV-specific IgA obtained on the same specimens were compared to PCR results. PCR had a specificity of 100% among 83 specimens from 50 babies known to be uninfected. Sensitivity among 26 HIV-infected infants older than 3 months was 98% (44 of 45 specimens); the one negative specimen, which had also been culture negative, gave a positive PCR result on the remaining aliquot when tested after decoding. Among infected infants less than 3 months old, which is an age when diagnosis by other assays is most problematic, PCR identified 10 of 10 patients (10 of 11 specimens) including two younger than one month. Viral culture showed the best concordance with PCR; however, in three infants, positive PCR results were observed several months before positive results were observed by viral culture.