Non-alcoholic fatty liver disease (NAFLD) is a wide-spread chronic liver condition that places patients at risk of developing cardiovascular diseases and may progress to cirrhosis or hepatocellular carcinoma if untreated. Challenges in clinical and basic research are caused by poor understanding of NAFLD mechanisms. The purpose of current study is to describe molecular changes occurring in human liver during NAFLD progression by defining a reproducible gene expression signature. We conduct a systematic meta-analysis of published human gene expression studies on liver biopsies and bariatric surgery samples of NAFLD patients. We relate gene expression levels with histology scores using regression models and identify a set of genes showing consistent-sign associations with NAFLD progression that are replicated in at least three independent studies. The analysis reveals genes that have not been previously characterized in the context of NAFLD such as HORMAD2 and LINC01554. In addition, we highlight biomarker opportunities for risk stratification and known drugs that could be used as tool compounds to study NAFLD in model systems. We identify gaps in current knowledge of molecular mechanisms of NAFLD progression and discuss ways to address them. Finally, we provide an extensive data supplement containing meta-analysis results in a computer-readable format.