Genotyping by high-resolution melting analysis of small amplicons is homogeneous and simple. However, this approach can be limited by physical and chemical components of the system that contribute to intersample melting variation. It is challenging for this method to distinguish homozygous G::C from C::G or A::T from T::A base-pair neutral variants, which comprise ∼16% of all human single nucleotide polymorphisms (SNPs). We used internal oligonucleotide calibrators and custom analysis software to improve small amplicon (42–86 bp) genotyping on the LightScanner®. Three G/C (PAH c.1155C>G, CHK2 c.1-3850G>C and candidate gene BX647987 c.261+22,290C>G) and three T/A (CPS1 c.3405-29A>T, OTC c.299-8T>A and MSH2 c.1511-9A>T) human single nucleotide variants were analyzed. Calibration improved homozygote genotyping accuracy from 91.7 to 99.7% across 1105 amplicons from 141 samples for five of the six targets. The average Tm standard deviations of these targets decreased from 0.067°C before calibration to 0.022°C after calibration. We were unable to generate a small amplicon that could discriminate the BX647987 c.261+22,290C>G (rs1869458) SNP, despite reducing standard deviations from 0.086°C to 0.032°C. Two of the sites contained symmetric nearest neighbors adjacent to the SNPs. Unexpectedly, we were able to distinguish these homozygotes by Tm even though current nearest neighbor models predict that the two homozygous alleles would be identical.