Background It remains unclear whether hepatectomy for hepatocellular carcinoma should be performed as an anatomic resection (AR) or a nonanatomic resection (NAR). Because no randomized controlled trials are currently available on this topic, a meta-regression analysis was performed on available observational studies to control for confounding variables. Methods A systematic review of studies published from 1990 to 2011 in the PubMed and Embase databases was performed. Patient and disease-free survival (DFS), postoperative mortality, and morbidity were considered as outcomes. Results are expressed as relative risk (RR) or weighted mean differences with 95 % of confidence interval. Results Eighteen observational studies involving 9,036 patients were analyzed: 4,012 were in the AR group and 5,024 in the NAR group. Meta-analysis suggested that AR provided better 5-year patient survival (RR 1.14; P  = 0.001) and DFS than NAR (RR 1.38; P  = 0.001). However, patients in the NAR group were characterized by a higher prevalence of cirrhosis (RR 1.27; P  = 0.010), more advanced hepatic dysfunction (RR 0.90 for Child-Pugh class A; P  = 0.001) and smaller tumor size (weighted mean difference 0.36 cm; P  < 0.001) compared with patients in the AR group. Meta-regression analysis showed that the different proportion of cirrhosis in the NAR group significantly affected both 5-year patient survival (RR 1.28; P  = 0.016) and DFS (RR 1.74; P  = 0.022). Tumor size only slightly affected DFS (RR 1.72; P  = 0.076). Postoperative mortality and morbidity were unaffected ( P  > 0.05 in all cases). Conclusions Patient survival and DFS after AR seem to be superior to NAR because the worse liver function reserve in the NAR group significantly affects prognosis.