Background: Hepatocellular carcinoma (HCC) is one of the most common cancers in China, and one of the leading causes of cancer-related deaths in the country. With a 5-year survival rate of only 15-20%, early detection is crucial to improve the treatment and survival of HCC patients. Currently, alpha-fetoprotein (AFP) is commonly used as a serum marker for HCC, but it is not a sufficiently specific and can cause false positive readings due to elevated levels caused by other liver conditions. An alternative method is to use circulating free DNA (cfDNA) released by tumor cells as cancer-screening targets, which has been shown to be a more sensitive and specific biomarkers for HCC detection. This study aims to develop a non-invasive screening assay based on cfDNA features to improve the detection of early-stage HCC. Methods: Candidate methylation markers for HCC detection were collected and evaluated using GEO, TCGA and in-house datasets, 1601 of which were incorporated into a targeted sequencing panel named HcSeer. Multiple types of cfDNA features were constructed from the sequencing data, which included methylation-related features such as methylation haplotype blocks (MHBs) and methylated haplotype fraction (MHF), and fragmentomics features such as end motif and CNV. For model building, Cancer and healthy plasma samples were randomly divided into a training and a testing set at a 2:1 ratio. A two-step deep neural network model was built to classify HCC using selected features of both types. Results: We previously enrolled a total of 401 plasma samples (200 healthy, 201 HCC) for model construction and the performance of the HcSeer model have been documented. An independent validation cohort of 421 plasma samples (280 healthy, 141 HCC) was currently collected from different centers. In this independent validation, the HcSeer model achieved an AUC of 0.98 with a sensitivity of 96.5% at a specificity of 96.4%. Importantly, HcSeer maintained a high sensitivity for HCC across all stages: 94.3%, 96%, 100% and 100% for stage I – IV cases, respectively. When compared to AFP, HcSeer achieved a significantly higher sensitivity of 94% than AFP’s 55% in 137 HCC cases having AFP level tested. When AFP level was combined with the HcSeer model, the sensitivity for HCC further increased to 96%. Conclusions: This study demonstrated that the DNA methylation and fragmentomics patterns of cfDNA can accurately distinguish HCC and healthy plasma samples, particularly in the early stages of HCC. The combination of the HcSeer and AFP further improved the accuracy of the prediction. Although this study was limited in sample size, it clearly showed the potential of the HcSeer assay for accurate HCC detection in blood.