Prediction of hepatocellular carcinoma (HCC) risk becomes increasingly important with recently emerging HCC-predisposing conditions, namely non-alcoholic fatty liver disease and cured hepatitis C virus infection. These etiologies are accompanied with a relatively low HCC incidence rate (~1% per year or less), while affecting a large patient population. Hepatitis B virus infection remains a major HCC risk factor, but a majority of the patients are now on antiviral therapy, which substantially lowers, but does not eliminate, HCC risk. Thus, it is critically important to identify a small subset of patients who have elevated likelihood of developing HCC, to optimize the allocation of limited HCC screening resources to those who need it most and enable cost-effective early HCC diagnosis to prolong patient survival. To date, numerous clinical-variable-based HCC risk scores have been developed for specific clinical contexts defined by liver disease etiology, severity, and other factors. In parallel, various molecular features have been reported as potential HCC risk biomarkers, utilizing both tissue and body-fluid specimens. Deep-learning-based risk modeling is an emerging strategy. Although none of them has been widely incorporated in clinical care of liver disease patients yet, some have been undergoing the process of validation and clinical development. In this review, these risk scores and biomarker candidates are overviewed, and strategic issues in their validation and clinical translation are discussed.