To systematically identify published, peer-reviewed economic evaluations of treatments for advanced or metastatic hepatocellular carcinoma (HCC), and to assess and summarize previous modelling approaches, model specifications, data sources, and variation in model outcomes. Comprehensive searches were conducted in major electronic databases to identify economic evaluations and were supplemented by searches in conference proceedings and HTA bodies on 6 March 2018. No geographic or publication date limits were applied but reviews, editorials and non-English language publications were excluded. Twenty-two economic models described in 33 publications (12 full-text journal publications, 15 HTA documents and six conference abstracts) were included in data extraction. Sorafenib monotherapy was the most commonly investigated therapy (in 15 studies, and was compared to best supportive care (BSC) in 10 studies), followed by regorafenib (investigated in three studies). The majority of the evaluations were performed using Markov modelling (17 studies), although two recent studies used a partitioned-survival model. The Markov models used either three or four health states and clinical efficacy data (e.g., time to progression, overall survival, progression-free survival) were mostly derived from pivotal randomised clinical trials. Half of the included models incorporated adverse events and health-related quality of life data. Across studies, the reported incremental cost-effectiveness ratio was generally most sensitive to survival estimates (overall survival and time to progression), utility values, BSC costs and treatment costs. Despite of similarities in clinical inputs and model structures, the heterogeneity due to country-specific settings and the year of analysis made cross-study comparisons inappropriate. Moderate number of economic studies is currently available. However, further efforts would be required to truly understand the relative cost-effectiveness of therapies for patients with advanced metastatic HCC. In particular, standardization of model structure, outcomes and input parameters would be needed to compare across evaluations.