OBJECTIVE Identifying patients with non-alcoholic steatohepatitis (NASH), a more aggressive form with a worse prognosis than for simple steatosis, is highly important. Liver biopsy still remains the gold standard for diagnosing NASH, but with limitations. The diagnostic value of serum cytokeratin-18 (CK-18) in predicting NASH is still indefinite. METHODS We selected relevant studies by a literature search of the PubMed, Ovid Medline and Cochrane Library databases up to January 2012. A DerSimonian-Laird random effects model was used to compute the pooled estimates of sensitivity (SEN), specificity (SPE), and diagnostic odds ratio (DOR), and a summary receiver operating characteristic (SROC) curve was constructed. Stratified analysis was performed to explore the heterogeneity in test accuracy. Funnel plot and Egger's regression were performed to assess publication bias. RESULTS A total of 10 studies with 838 patients were included (nine CK-18 fragments and five total CK-18 studies) in this meta-analysis. Among nine CK-18 fragment studies with a significant publication bias, the pooled results on SEN, SPE and DOR were 0.83 (95% CI, 0.80-0.86), 0.71 (95% CI, 0.66-0.76) and 11.90 (95% CI, 6.05-23.40), respectively, and age and body mass index were most strongly associated with the observed heterogeneity. Among five total CK-18 studies with homogeneity, the pooled results of SEN, SPE and DOR were 0.77% (95% CI, 0.70-0.83), 0.71 (95% CI, 0.65-0.77) and 7.99 (95% CI, 4.09-15.62), respectively. The area under the ROC curve (± SE) of CK-18 fragments and total CK-18 were 0.8445 (± 0.0306) and 0.8170 (± 0.0429), respectively. CONCLUSIONS Both CK-18 fragments and total CK-18 have a clinically meaningful benefit in noninvasive diagnosing of NASH, though total CK-18 has a relatively low diagnostic accuracy. CK-18 fragments may be a useful biomarker for screening rather than identifying NASH.