BACKGROUND: Blood-based analysis of circulating tumor DNA (ctDNA) is a promising tool for cancer screening, monitoring relapse/recurrence and evaluating response to treatment. Although plasma is widely used to obtain ctDNA, biorepositories worldwide possess a huge number of serum samples and comparative studies on the use of serum vs plasma as ctDNA sources are essential. METHODS: We analyzed cell-free DNA (cfDNA) from matched EDTA-plasma and serum samples from healthy donors and patients with colorectal or lung cancer, and used targeted next-generation sequencing to evaluate mutation detection efficiency and reproducibility. Matched samples from healthy individuals were spiked with reference oligonucleotides and sequenced using the Ion-S5 Oncomine-Pan-Cancer panel. Detection efficiency in matched samples from patients with cancer was evaluated using 2 distinct gene panels and compared to mutations found in tissue-biopsy samples at diagnosis. RESULTS: Mean total cfDNA was 55% higher in serum samples and the presence of longer DNA fragments was significantly increased in serum compared with plasma samples (P 0.0001 to 0.015). Spiked mutated nucleotides were detected in both samples, but allele frequencies (AF) were approximately half in serum compared with plasma, suggesting ctDNA from serum was more diluted by DNA of noncancerous origins. Matched samples from patients with cancer revealed that up to 44.8% of mutations with low AF were missed in serum samples and concordance rates with somatic mutations found in tissue biopsy at diagnosis was better in plasma samples. CONCLUSION: The use of serum in retrospective studies should consider the limitations for detecting low AF mutations. Plasma is clearly preferable for prospective clinical applications of liquid biopsy.