Objective: Studies have predominantly focused on progression from mild cognitive impairment (MCI) to dementia, but little is known about individuals who remain as stable MCI across time. This study sought to examine whether baseline cognitive performance predicts MCI stability at 3-year follow-up. Method: Baseline cognitive data from the National Alzheimer's Coordinating Center were examined for 1, 059 subjects (556 males, 503 females, mean age = 74.73 ± 8.73, mean education = 14.93 ± 3.46) diagnosed with one of the four MCI subtypes (amnestic single-domain = 267, amnestic multiple-domain = 208, non-amnestic single-domain = 74, non-amnestic multiple-domain = 47) at initial visit who had sequential 3-year follow-ups. Subjects were classified as stable MCI if they were diagnosed at every follow-up visit as MCI, regardless of subtype. Subjects were classified as progressors if they were diagnosed at both 2-year and 3-year follow-up visits as demented, regardless of the diagnosis at 1-year. Results: At 3 years, 596 subjects (56%) remained stable MCI, while 463 (44%) progressed to dementia. Logistic regression was used to predict group membership (stable versus progression) with baseline cognitive test scores in memory, language, attention, processing speed, and executive functioning domains entered as predictors. Higher scores on story delayed recall (χ2 = 122.67, OR = 1.19, p < .001), digit symbol (χ2 = 24.69, OR = 1.24, p < .001), and category fluency (animals: χ2 = 10.86, OR = 1.27, p = .001; vegetables: χ2 = 4.75 OR = 1.19, p = .029), were predictors of stable MCI. Conclusion(s): MCI patients with higher recall, processing speed, and category fluency test scores were up to 1.3 times more likely to remain diagnostically stable at 3-years, suggesting that these tests may predict subsequent stable course of MCI.