Background: “Deep” sequencing allows detection of low frequency minority variants, but establishing the analytical characteristics of such assays, including the error rate and limit of detection, will require repeated sequencing of known reference samples. We present sequencing results from three next-generation platforms for one such potential sample. Methodology: Fifteen infectious viral clones, each containing a single NNRTI resistance mutation (V90I, K101E, K103N, V108I, E138A/G/K/Q, V179D, Y181C, Y188C, G190A/S, M230L and P236L) were created by site directed mutagenesis of HIV-1LAI. Mutants and wild-type viruses were pooled at equal infectivity into a single sample, “RSVP” (reference sample for validating platforms), which was extracted and subjected to RT-PCR. Two second round PCR amplicons were generated: a 1.8 kb region of pol (HXB2 2469-4295) sequenced on the PacBioR RS (v1), and 339 bp region of RT (HXB2 2813-3151) was sequenced on the Illumina MiSeq (v2 2×250 paired end) and 454 Life Sciences GS-Jr. Cross-platform variation was evaluated including amino acid (aa) frequency of the expected NNRTI mutants and error rates in the region covered by all three platforms (RT codons 96-194). A logistic regression model was used to model the error rates. In addition, serial dilutions of the 1st round PCR product were sequenced by MiSeq. Results: Approximately 50,000, 185,000, and 3100 sequence reads were obtained for PacBio, MiSeq and 454 respectively. The frequency of minor NNRTI resistance variants detected by the three platforms was similar (see figure). When raw sequences were analyzed (no quality score filtering) marginally higher error rates were observed for PacBio compared to MiSeq and 454. For MiSeq, discarding bases with quality below a specific cutoff (q-score 10, 15, 20, 25) resulted in improved error rates, with q20 balancing a tradeoff between read retention and sequencing accuracy (q20 error rate 1.2%). Sequencing errors in homopolymer-rich regions were frequently observed in 454 sequences, occasionally in PacBio, but rarely in MiSeq. Additionally, MiSeq results were robust to dilution of the 1st round PCR product up to 200,000-fold. Conclusions: We have characterized the performance of three next-generation sequencing platforms using a proposed HIV reference sample; RSVP is a mixture of protease-RT from 16 HIV-1 variants that we are making available to interested groups to use when validating the detection of minority NNRTI resistant variants.