Nucleotide sequences in three hypervariable regions of the human immunodeficiency virus type 1 (HIV-1) env gene were obtained by sequencing provirus present in peripheral blood mononuclear cells of HIV-infected individuals. Single molecules of target sequences were isolated by limiting dilution and amplified in two stages by the polymerase chain reaction, using nested primers. The product was directly sequenced to avoid errors introduced by Taq polymerase during the amplification process. There was extensive variation between sequences from the same individual as well as between sequences from different individuals. Interpatient variability was markedly less in individuals infected from a common source. A high proportion of amino acid substitutions in the hypervariable regions altered the number and positions of potential N-linked glycosylation sites. Sequences in two hypervariable regions frequently contained short (3- to 15-bp) duplications or deletions, and by amplifying peripheral blood mononuclear cell DNA containing 10(2) or 10(3) proviral molecules and analyzing the product by high-resolution electrophoresis, the total number and abundance of distinct length variants within an individual could be estimated, providing a more comprehensive analysis of the variants present than would be obtained by sequencing alone. Sequences from many individuals showed frequent amino acid substitutions at certain key positions for neutralizing-antibody and cytotoxic T-cell recognition in the immunodominant loop. The rates of synonymous and nonsynonymous nucleotide substitution in the region of this and flanking regions indicate that strong positive selection for amino acid change is operating in the generation of antigenic diversity.