In vitro animal and human models were used to evaluate the potential of chemical shift magnetic resonance imaging (MRI) for assessing fatty liver. Phantoms of varying fat content were created from mayonnaise-agar preparations. Fatty liver was induced in eight rats by feeding them ethanol for three to six weeks (36% of total calories), whereas eight control rats were fed a normal diet. T1-weighted in-phase and opposed-phase MR images were obtained of the phantoms animals, and 28 human subjects. Additional images obtained in animals included long TR images with in-phase and opposed-phase technique, and hybrid chemical shift water and fat suppression. The rats were killed and histologic status was graded blindly by a hepatopathologist as normal, mild, moderate, or severe fatty change, for correlation with MR grading. Quantitative analysis of MR images included fat signal fraction for animals, and relative signal decrease between in-phase and opposed-phase images for phantom and human data. Phantom in-phase signal increased linearly with respect to fat content, whereas opposed-phase signal decreased linearly. MRI and histologic grading of rat livers were highly correlated, especially when based on water suppression images (r = 0.91, P = .0001). Opposed-phase images were also highly correlated, while fat suppression images were less effective. There was no overlap between MR-derived fat fractions for control (2.6%-5.7%) versus ethanol-fed rats (7.7%-17.9%, P = .0002). Human liver considered to be fatty by visual inspection (n = 8) had higher relative signal decrease than nonfatty liver (n = 22) (P less than .001). Phantom, animal, and human data demonstrate that comparison of T1-weighted in-phase and opposed-phase images is both practical and sensitive in the detection and grading of fatty liver.