Kinetics of incorporation of correct and incorrect deoxynucleotides by three reverse transcriptases have been followed, by gel assay, on a series of DNA templates, including part of the HIV-1 gag DNA minus strand. Insertion kinetics for the properly matched nucleotide at a given place on the template vary strongly from one enzyme to the next. No significant correlation is found between the site-specific Michaelis constants, while the maximal velocities are more closely connected. For a given reverse transcriptase these parameters are strongly influenced by the DNA sequence. A systematic evaluation of the frequencies of misincorporation was then performed at 46 positions. Again great variability was found, precluding a very accurate evaluation of an average misincorporation frequency for a given enzyme and a given mismatch. Qualitatively however, HIV-1 reverse transcriptase is certainly not more error-prone in this assay than the other enzymes assayed. The patterns of misincorporations were again very dependent on the enzyme used to replicate a given template. The variability of the gag sequence observed in vivo among various HIV-1 isolates was compared with the patterns of misincorporations obtained in vitro on the same sequence with HIV-1, AMV and MoMLV reverse transcriptases. A fair agreement was found with the pattern observed in the polymerization directed by the HIV-1 reverse transcriptase. The correlation is less important in the two other cases. However some specific changes observed in vivo cannot be accounted for by our misincorporation assay, even when performed with the homologous enzyme, suggesting that an important class of mismatches can only be generated during reverse transcription of the RNA strand. Additional data, using a complementary DNA (positive) strand as a gag template support this hypothesis.