Identification of potentially modifiable risk factors for cognitive deterioration is important. We conducted a prospective study of 5,607 subjects with normal cognition and 2,500 subjects with mild cognitive impairment (MCI) at 30 Alzheimer's Disease Centers in the Unites States between 2005 and 2011. Cox regression was used to determine whether depression predicted transition from normal to MCI, or MCI to Alzheimer's disease (AD). Over an average of 3.3 visits, 15% of normal subjects transitioned to MCI (62/1000 per year), while 38% of MCI subjects transitioned to AD (146/1000 per year). At baseline, 22% of participants had recent (within the last two years) depression defined by clinician judgment; 9% and 17% were depressed using the Geriatric Depression Scale (GDS score ≥5) and the Neuropsychiatric Inventory Questionnaire (NPI-Q), respectively. At baseline, depressed subjects performed significantly worse on cognitive tests. Those always depressed throughout follow-up had an increased risk for progression from normal to MCI (RR = 2.35; 95% CI 1.93-3.08) versus never depressed. Normal subjects, identified as depressed at first visit but subsequently improved, were found to have lower risk of progression (RR 1.40 (1.01-1.95)). The 'always depressed' had only a modest increased risk of progression from MCI to AD (RR = 1.21 (1.00-1.46). Results were similar using time-dependent variables for depression or when defining depression via the GDS or NPI-Q. We found no effect of earlier depression (>2 years past). The effect of recent depression did not differ by antidepressant treatment, APOE4 allele status, or type of MCI. In conclusion, late-life depression is a strong risk factor for normal subjects progressing to MCI.