Metastatic colorectal cancer (mCRC) is frequently characterized by the presence of mutations of the KRAS oncogene, which are generally associated with a poor response to treatment with anti-epidermal growth factor receptor (anti-EGFR) monoclonal antibodies. With the methods currently used, a case is classified as KRAS-mutated when approximately 20% of the cells bear an activating KRAS mutation. These considerations raise the question of whether cells with a mutated KRAS can be found in mCRC cases classified as KRAS wild-type when more sensitive methods are used. In addition, the issue arises of whether these mCRC cases with low proportion of KRAS-mutated cells could account at least in part for the therapeutic failure of anti-EGFR therapies that occur in 40-60% of cases classified as KRAS wild type. In this study, we compared the classical assays with a very sensitive test, a locked nucleic acid (LNA) polymerase chain reaction (PCR), capable of detecting KRAS-mutated alleles at extremely low frequency (detection sensitivity limit 0.25% mutated DNA/wild-type DNA). By analyzing a cohort of 213 mCRC patients for KRAS mutations, we found a 20.6% discordance between the sequencing/TheraScreen methods and the LNA-PCR. Indeed, 44 mCRC patients initially considered KRAS wild type were reclassified as KRAS mutated by using the LNA-PCR test. These patients were more numerous among individuals displaying a clinical failure to anti-EGFR therapies. Failure to respond to these biological treatments occurred even in the absence of mutations in other EGFR pathway components such as BRAF.