Background The guidelines recommend surveillance with liver ultrasound (US) for early detection of hepatocellular carcinoma (HCC). The accuracy of US is operator-dependent and is limited in those with severe obesity or ascites. Magnetic resonance imaging (MRI) may have higher accuracy for detecting HCC, especially early HCC. Objective To assess the accuracy of US for HCC detection using MRI as the clinical gold standard. Methods We performed a retrospective study of 223 patients with cirrhosis who had both abdominal US and MRI within 6 months of each other, between October 2013 and November 2018. Using MRI as the gold standard for detecting HCC, the test characteristics of US and concordance with MRI were evaluated, and potential patient characteristics associated with the false-negative US were examined. Results Of 223 patients, MRI detected HCC in 70 patients while US detected HCC only in 38; thus, the overall sensitivity of US was 54% in detecting HCC, using MRI as the reference standard. The prevalence-adjusted and bias-adjusted kappa (PABAK) was 0.65 for the entire group. However, PABAK was only 0.04 when the subgroup with HCC (n = 70) was analysed. Higher Model for End-Stage Liver Disease (MELD) score was the only factor that was associated with decreased US sensitivity, with the odds of false-negative US increasing by 9.8% for each 1-point increase in MELD. Conclusion Our study showed that MRI performed better than the US in detecting HCC in patients with cirrhosis. Patients with higher MELD may require an MRI for HCC surveillance for early diagnosis of HCC.