Background Lung cancer is the leading cause of cancer morbidity and mortality worldwide, however, the individualized treatment is still unsatisfactory. DNA methylation can affect gene regulation and may be one of the most valuable biomarkers in predicting the prognosis of lung adenocarcinoma. This study was aimed to identify methylation CpG sites that may be used to predict lung adenocarcinoma prognosis. Methods The Cancer Genome Atlas (TCGA) database was used to detect methylation CpG sites associated with lung adenocarcinoma prognosis and construct a methylation signature model. Then, a Chinese cohort was carried out to estimate the association between methylation and lung adenocarcinoma prognosis. Biological function studies, including demethylation treatment, cell proliferative capacity, and gene expression changes in lung adenocarcinoma cell lines, were further performed. Results In the TCGA set, three methylation CpG sites were selected that were associated with lung adenocarcinoma prognosis (cg14517217, cg15386964, and cg18878992). The risk of mortality was increased in lung adenocarcinoma patients with the gradual increase level of methylation signature based on three methylation sites levels (HR = 45.30, 95% CI = 26.69-66.83; p < 0.001). The C-statistic value increased to 0.77 when age, gender, and other clinical variables were added to the signature to prediction model. A similar situation was confirmed in Chinese lung adenocarcinoma cohort. In the biological function studies, the proliferative capacity of cell lines was inhibited when the cells were demethylated with 5-aza-2'-deoxycytidine (5-aza-2dC). The mRNA and protein expression levels of SEPT9 and HIST1H2BH (cg14517217 and cg15386964) were downregulated with different concentrations of 5-aza-2dC treatment, while cg18878992 showed the opposite result. Conclusion This study is the first to develop a three-CpG-based model for lung adenocarcinoma, which is a practical and useful tool for prognostic prediction that has been validated in a Chinese population.