DNA samples from formalin-fixed paraffin-embedded tissues are highly degraded with variable quality, and this imposes a big challenge for targeted sequencing due to false positives, largely caused by PCR errors and cytosine deamination. To eliminate false positives, a common practice is to validate the detected variants by Sanger sequencing or perform targeted sequencing in duplicate. Technically, PCR errors could be removed by molecular barcoding of template DNA prior to amplification as in the HaloPlexHS design. Nonetheless, it is uncertain to what extent variants detected using this approach should be further validated. Here, we addressed this question by correlating variant reproducibility with DNA quality using HaloPlexHS target enrichment and Illumina HiSeq4000, together with an in-house validated variant calling algorithm. The overall sequencing coverage, as shown by analyses of 70 genes in 266 cases of large B-cell lymphoma, was excellent (98%) in DNA samples amenable for PCR of ≥400 bp, but suboptimal (92%) and poor (80%) in those amenable for PCR of 300 bp and 200 bp respectively. By mutation analysis in duplicate in 93 cases, we demonstrated that 20 alternative allele depth (AAD) was an optimal cut-off value for separating reproducible from non-reproducible variants in DNA samples amenable for PCR of ≥300 bp, with 97% sensitivity and 100% specificity. By cross validation with a previously established targeted sequencing protocol by Fluidigm-PCR and Illumina MiSeq, the HaloPlexHS protocol was shown to be highly sensitive and specific in mutation screening. To conclude, we proposed a stratified approach for mutation screening by HaloplexHS and Illumina HiSeq4000 according to DNA quality. DNA samples with good quality (≥400 bp) are amenable for mutation analysis with a single replicate, with only variants at 15-20 AAD requiring for further validation, while those with suboptimal quality (300 bp) are better analysed in duplicate with reproducible variants at >15 AAD regarded as true genetic changes.