Nonalcoholic fatty liver disease (NAFLD) has become a common health care burden worldwide. The high heterogeneity of NAFLD remains elusive and impairs outcomes of clinical diagnosis and pharmacotherapy. Several NAFLD classifications have been proposed on the basis of clinical, genetic, alcoholic, or serum metabolic analyses. Yet, accurately predicting the progression of NAFLD to cirrhosis or hepatocellular carcinoma (HCC) in patients remains a challenge. Here, on the basis of a Chinese cohort of patients, we classified NAFLD into three distinct molecular subtypes (NAFLD-mSI, NAFLD-mSII, and NAFLD-mSIII) using integrative multiomics including whole-genome sequencing (WGS), proteomics, phosphoproteomics, lipidomics, and metabolomics across a broad range of liver, blood, and urine specimens. We found that NAFLD-mSI had higher expression of CYP1A2 and CYP3A4, which alleviate hepatic steatosis through mediating free fatty acid/bile acid–mTOR–FXR/PPARα signaling. NAFLD-mSII displayed an elevated risk of liver cirrhosis along with increased hepatic infiltration of M1 and M2 macrophages because of lipid-triggered hepatic CCL2 and CRP production. NAFLD-mSIII exhibited a potential risk for HCC development by increased transcription of CEBPB- and ERCC3-regulated oncogenes because of activation of the EGF-EGFR/CHKA/PI3K-PDK1-AKT cascade. Next, we validated the existence of these three NAFLD molecular subtypes in an external cohort comprising 92 patients with NAFLD across three different Chinese hospitals. These findings may aid in understanding the molecular features underlying NAFLD heterogeneity, thereby facilitating clinical diagnosis and treatment strategies with the aim of preventing the development of liver cirrhosis and HCC.