Detection of activating EGFR mutations in NSCLC is the prerequisite for individualised therapy with receptor tyrosine kinase inhibitors (TKI). In contrast, mutant downstream effector KRAS is associated with TKI resistance. Accordingly, EGFR mutation status is routinely examined in NSCLC specimens, but the employed methods may have a dramatic impact on the interpretation of results and, consequently, therapeutic decisions. Specimens with low tumour cell content are at particular risk for false-negative EGFR mutation reporting by sequencing with Sanger chemistry. To improve reliability of detecting clinically relevant mutant variants of EGFR and KRAS, we took full advantage of 454 deep sequencing and developed a two-step amplification protocol for the analysis of EGFR exons 18–21 and KRAS exons 2 and 3. We systematically addressed the sensitivity, reproducibility and specificity of the developed assay. Mutations could be reliably identified down to an allele frequency of 0.2–1.5 %, as opposed to 10–20 % detection limit of Sanger sequencing. High reproducibility (0–2.1 % variant frequency) and very low background level (0.4–0.8 % frequency) further complement the reliability of this assay. Notably, re-evaluation of 16 NSCLC samples with low tumour cell content ≤40 % and EGFR wild type status according to Sanger sequencing revealed clinically relevant EGFR mutations at allele frequencies of 0.9–10 % in seven cases. In summary, this novel two-step amplification protocol with 454 deep sequencing is superior to Sanger sequencing with significantly increased sensitivity, enabling reliable analysis of EGFR and KRAS in NSCLC samples independent of the tumour cell content.