Technical Abstract: Discovery of rare mutations in populations requires methods for processing and analyzing in parallel many individuals. Previous TILLING methods employed enzymatic or physical discrimination of heteroduplexed from homoduplexed target DNA. We used mutant populations of rice and wheat to develop a method based on Illumina sequencing of gene fragments amplified from multi-dimensionally pooled templates. We implemented sequence spiking for sample verification and sequence barcoding for multiplex lane loading. Changes collected after alignment of short reads to the reference were evaluated for sequencing quality, for occurrence in multiple pools, and for statistical relevance producing a Bayesian score with an associated confidence threshold. Errors were favored by varying coverage and sequence context. While these would be problematic in a mono-dimensional assay, they could be addressed by the combination of concurrent discovery in two or three pools with the bioinformatic treatment, resulting in accurate discovery.