Next‐generation sequencing of reduced‐representation genomic libraries provides a powerful methodology for genotyping thousands of single‐nucleotide polymorphisms (SNPs) among individuals of nonmodel species. Utilizing genotype data in the absence of a reference genome, however, presents a number of challenges. One major challenge is the trade‐off between splitting alleles at a single locus into separate clusters (loci), creating inflated homozygosity, and lumping multiple loci into a single contig (locus), creating artefacts and inflated heterozygosity. This issue has been addressed primarily through the use of similarity cut‐offs in sequence clustering. Here, two commonly employed, postclustering filtering methods (read depth and excess heterozygosity) used to identify incorrectly assembled loci are compared with haplotyping, another postclustering filtering approach. Simulated and empirical data sets were used to demonstrate that each of the three methods separately identified incorrectly assembled loci; more optimal results were achieved when the three methods were applied in combination. The results confirmed that including incorrectly assembled loci in population‐genetic data sets inflates estimates of heterozygosity and deflates estimates of population divergence. Additionally, at low levels of population divergence, physical linkage between SNPs within a locus created artificial clustering in analyses that assume markers are independent. Haplotyping SNPs within a locus effectively neutralized the physical linkage issue without having to thin data to a single SNP per locus. We introduce a Perl script that haplotypes polymorphisms, using data from single or paired‐end reads, and identifies potentially problematic loci.