Objectives: To describe magnetic resonance imaging characteristics in a large sample of subjects with mild cognitive impairment (MCI) and to investigate associations between these characteristics and cognition. Design: Cohort study. Setting: Baseline data of a randomized, double-blind, placebo-controlled clinical trial of galantamine in MCI (Trial Registration: clinicaltrials.gov Identifier: NCT00236574). Patients: Included in the study were 896 subjects with MCI (age [mean ± SD], 70 ± 9 years; 54% women) with available clinical and magnetic resonance imaging data. Main Outcome Measures: Neuropsychology: Alzheimer Disease Assessment Scale, cognitive subscale, MCI version, assessing global cognition; delayed recall on the New York University Paragraph Recall Test, assessing episodic memory; and Digit Symbol Substitution Test, assessing executive function. Neuroimaging: Medial Temporal Lobe Atrophy (MTA) Rating Scale (0-4) and Age-Related White Matter Changes Scale (0-30), assessing white matter hyperintensities(WMHs); and lacune counts. Results: Median MTA score was 2 (range, 0-4), and mean (±SD) Age-Related White Matter Changes Scale score 6.0 (±4.7). Lacunes were present in 33% of subjects. In unadjusted models, increasing MTA and WMHs were associated with poorer performance on all cognitive tests, and lacunes with poorer performance on the Alzheimer Disease Assessment Scale, cognitive subscale, MCI version, and the Digit Symbol Substitution Test. In multi-variable models, including magnetic resonance imaging measures simultaneously, MTA remained a predictor of cognition, whereas WMH had no independent predictive value. There was an interaction between MTA and lacunes: the strength of the association with the Digit Symbol Substitution Test increased with decreasing MTA. Conclusions: Medial temporal lobe atrophy seems to be a more important predictor of cognition than small-vessel disease in MCI. Lacunes were associated with performance on the Digit Symbol Substitution Test, especially in subjects with milder MTA. Although WMHs were prevalent and associated with cognition in unadjusted analyses, there was no discernible association between WMHs and the cognitive measures in this study after adjustment for age.