Background: Ovarian cancer is the leading cause of gynecological cancer-related mortality. The majority of patients have poor prognosis due to the lack of available biomarkers capable of predicting prognosis accurately and providing novel targeted therapy options.Methods: The methylation array data, gene expression profile and clinical information of 351 ovarian serous cystadenocarcinoma (OSC) samples as well as the methylation array data of another 10 normal ovarian epithelial tissues were accessed from The Cancer Genome Atlas (TCGA) and served as training dataset. An algorithm, MethylMix, was on the basis of a β mixture model to compare the DNA methylation status in tumor to normal DNA methylation status and then screened out hypo and hypermethylated genes of OSC, which were transcriptionally predictive. Gene Ontology (GO) enrichment and ConsensusPathDB pathway analysis of differential methylation-driven genes were further performed. A linear risk model was constructed through the univariate and multivariate COX regression with Akaike Information Criterion (AIC). Datasets of OSC from Gene Expression Omnibus (GEO) database were served as validation datasets to determine the prognostic value of the risk model. The Kaplan-Meier analysis was then used to evaluate the ability of the risk model to distinguish the survival in training dataset or validation dataset, even in subgroups of patients with different FIGO stage, histologic grade, residual tumor after surgery, and anatomic neoplasm subdivision. The receiver operating characteristic (ROC) curve was also applied to determine the value of risk model on predicting the prognostic survival of patients as well as compared with other 15 known prognostic biomarkers in OSC.Results: Totally 171 differential methylation-driven genes were identified between OSC and normal samples. Five of them, UBB, PLAT, TMOD1, KCNJ11, and CDSN, were selected to construct the prognostic risk model. The Kaplan-Meier analysis demonstrated that the survival rate in low-risk cohort was significantly higher than that in high-risk cohort in both training dataset and validation dataset (P<0.0001 for training dataset; P<0.05 for validation dataset). Further, the ROC analysis confirmed the prognosis-predict value of the risk model applicable not only in different group of patients but also more accurately when compared with other biomarkers.Conclusions: Our study successfully identified multiple methylation-driven genes in OSC and constructed a novel prognostic risk model to provide bioinformatic basis and an important tool for guiding subsequent OSC early diagnosis, prognosis assessment, and clinical treatment.