Background: Tumor-derived cfDNA fragments are observed to be shorter in length than normal cfDNA. This size (length) difference can be analyzed as a tumor-specific signal. Whole genome or probe hybridization-based NGS methods can capture cfDNA fragments of native sizes. However, amplicon-based NGS assays are not directly amenable to cfDNA size analysis due to predetermination of amplicon sizes by design. Here we present a method to extract relative distribution of cfDNA fragment lengths from a targeted amplicon-based assay and show its utility in cancer detection. Methods: The LiquidHALLMARK cfDNA assay is an amplicon-based NGS test for the sensitive detection of genomic alterations in 80 genes. Although panel design is optimized for cfDNA with average amplicon length of ̃150 bp, the consecutive tiling design of amplicons for genes with contiguously targeted regions, e.g. BRCA permits the formation of longer amplicons (> 150 bp) from physically subsequent primer pairs, provided longer template cfDNA molecules are present. Cancer samples (n = 281), clinically tested by LiquidHALLMARK during Sep 2020-Sep 2021, and healthy samples (n = 28) were included for analysis. For each sample, fragment lengths were inferred from sequencing alignment files, and binned into “short” (0-150 bp) and “long” (151-500 bp) groups. A relative “size ratio” of the total number of short vs. long fragments per sample was calculated, and examined with clinical features, plasma cfDNA concentration (cfDNA/ml) and highest mutation allele frequency (AF%), in a model to predict cancer. Results: Calculated size ratios (relative abundance of short fragments) were higher in cancer than normal samples (median 47.6 vs. 31.2, p < 0.001). Cholangiocarcinoma and colorectal cancer samples had the highest size ratios (medians: 62.9 and 60.9, respectively) in agreement with a genome-wide NGS study that profiled cfDNA sizes. Size ratios were higher in metastatic (n = 143) compared to early stage (n = 30) lung cancers (p = 0.0039), indicating a stage-dependent accumulation of shorter cfDNA fragments. Size ratio was correlated with cfDNA/ml (r= 0.63, p < 0.01) and AF% (r= 0.42, p < 0.01). In 20-fold cross-validation of a logistic regression model trained to predict cancer, average area under curve (AUC) was 0.82 using size ratio, 0.86 using cfDNA/ml, and increased to 0.95 with the two features combined. Conclusions: Our analysis shows that it is feasible to derive meaningful cfDNA fragment size information from amplicon-based NGS data. Importantly, relative fragment size distributions observed in cancer and healthy plasma samples by this method are concordant with alternate target capture methods. Fragment size ratios derived from relatively small, targeted amplicon panels are a novel feature that, combined with other molecular and clinical features, can enhance non-invasive methods of cancer detection.