The accurate detection of induced mutations is critical for both forward and reverse genetics studies. Experimental chemical mutagenesis induces relatively few single base changes per individual. In a complex eukaryotic genome, false positive detection of mutations can occur at or above this mutagenesis rate. We demonstrate here, using a population of ethyl methanesulfonate (EMS) treated Sorghum bicolor BTx623 individuals, that using replication to detect false positive induced variants in next-generation sequencing data permits higher throughput variant detection with greater accuracy. We used a lower sequence coverage depth (average of 7X) from 586 independently mutagenized individuals and detected 5,399,493 homozygous SNPs. Of these, 76% originated from only 57,872 genomic positions prone to false positive variant calling. These positions are characterized by high copy number paralogs where the error-prone SNP positions are at copies containing a variant at the SNP position. The ability of short stretches of homology to generate these error prone positions suggests that incompletely assembled or poorly mapped repeated sequences are one driver of these error prone positions.. Removal of these false positives left 1,275,872 homozygous and 477,531 heterozygous EMS-induced SNPs which, congruent with the mutagenic mechanism of EMS, were greater than 98% G:C to A:T transitions. Through this analysis we generated a database of sequence indexed mutants of Sorghum. This collection contains 4,035 high impact homozygous mutations in 3,637 genes and 56,514 homozygous missense mutations in 23,227 genes. Each line contains, on average, 2,177 annotated homozygous SNPs per genome, including seven likely gene knockouts and 96 missense mutations. The number of mutations in a transcript was linearly correlated with the transcript length and also the G+C count, but not with the GC/AT ratio. Analysis of the detected mutagenized positions identified CG-rich patches, and flanking sequences strongly influenced EMS-induced mutation rates. Our method for detecting false-positive induced mutations is generally applicable to any organism, is independent of the choice of in silico variant-calling algorithm, and is most valuable when the true mutation rate is likely to be low, such as in laboratory induced mutations or somatic mutation detection in medicine.