Background: Next-generation sequencing (NGS) can provide in-depth resolution of heterogeneous intra-patient viral populations, composed of one or more genetically distinct haplotypes. However, such analyses are complicated by small read length, up to two orders of magnitude shorter than the genome size, and by the error rate of amplification and sequencing. Identification of co-occurring resistance mutations, composition, and ongoing evolution of intrapatient viral populations on the basis of full-length genomes are of great interest, but remain unsolved. New approaches and tools are needed to reconstruct individual full-length haplotypes and to estimate their frequency in the population. Methodology: We combined five molecular HIV-1 strains to create a heterogeneous mixture of viral haplotypes. RNA was isolated and whole genomes amplified with three different protocols for the NGS technologies 454/Roche, Illumina, and PacBio. For each data set, we performed full-length haplotype reconstruction over the complete coding region of 8,627 bp, using the software packages QuasiRecomb and PredictHaplo. To determine the ground truth sequences and frequencies, we sequenced the HIV-1 strains individually and performed single-genome amplification (SGA) on the protease gene. Results: Applying the NGS platforms 454/Roche, Illumina, and PacBio, we amplified and sequenced 35, 5, and 11 amplicons, with average read length (mean}sd) 410}41 bp, 237}26 bp, and 1499}77 bp, and average coverage (range) 7,712 (2,237-22,661), 23,010 (5,084-47,232), and 1,741 (450-3,147) reads per nucleotide, respectively. We reconstructed all five full-length haplotypes error-free from p17 in gag to nef. Estimated frequencies were almost identical between the independent reconstruction methods QuasiRecomb and PredictHaplo, in agreement with SGA, and very stable for 100 bootstrap samples with a mean variance of 0.4% across strains and technologies. Full-length frequency estimates were more reliable than gene-wise estimates. Conclusions: Full-length haplotype reconstruction from a heterogeneous HIV-1 sample is technically feasible. It depends critically on amplicon layout and successful amplification, sufficient read length, and homogeneous coverage. Our protocols and methods to decompose a heterogeneous population into individual full-length haplotypes can be used to identify quasispecies structures and study evolution of viral populations.