In genome-wide association studies, high-level statistical analyses rely on the validity of the called genotypes, and different genotype calling algorithms (GCAs) have been proposed. We compared the GCAs Bayesian robust linear modeling using Mahalanobis distance (BRLMM), Chiamo++, and JAPL using the autosomal single-nucleotide polymorphisms (SNPs) from the 500 k Affymetrix Array Set data of the Framingham Heart Study as provided for the Genetic Analysis Workshop 16, Problem 2, and prepared standard quality control (sQC) for each algorithm. Using JAPL, most individuals were retained for the analysis. The lowest number of SNPs that successfully passed sQC was observed for BRLMM and the highest for Chiamo++. All three GCAs fulfilled all sQC criteria for 79% of the SNPs but at least one GCA failed for 18% of the SNPs. Previously undetected errors in strand coding were identified by comparing genotype concordances between GCAs. Concordance dropped with the number of GCAs failing sQC. We conclude that JAPL and Chiamo++ are the GCAs of choice if the aim is to keep as many subjects and SNPs as possible, respectively.