AIM: To compare ultrasound gradings of steatosis with fat fraction (FF) on magnetic resonance spectroscopy (MRS; the non-invasive reference standard for quantification of hepatic steatosis), and evaluate inter- and intraobserver variability in the ultrasound gradings. MATERIALS AND METHODS: Triple grading of hepatic ultrasound examination was performed by three independent graders on 131 people with type 2 diabetes. The stored images of 60 of these individuals were assessed twice by each grader on separate occasions. Fifty-eight patients were pre-selected on the basis of ultrasound grading (normal, indeterminate/mild steatosis, or severe steatosis) to undergo (1)H-MRS. The sensitivity and specificity of the ultrasound gradings were determined with reference to MRS data, using two cut-offs of FF to define steatosis, =9% and =6.1%. RESULTS: Median (intraquartile range) MRS FF (%) in the participants graded on ultrasound as normal, indeterminate/mild steatosis, and severe steatosis were 4.2 (1.2-5.7), 4.1 (3.1-8.5) and 19.4 (12.9-27.5), respectively. Using a liver FF of =6.1% on MRS to denote hepatic steatosis, the unadjusted sensitivity and specificity of ultrasound gradings (severe versus other grades of steatosis) were 71 and 100%, respectively. Interobserver agreement within one grade was observed in 79% of cases. Exact intraobserver agreement ranged from 62 to 87%. CONCLUSION: Hepatic ultrasound provided a good measure of the presence of significant hepatic steatosis with good intra- and interobserver agreement. The grading of a mildly steatotic liver was less secure and, in particular, there was considerable overlap in hepatic FF with those who had a normal liver on ultrasound.