Plasma amyloid-β (Aβ) levels have been proposed as biomarkers of Alzheimer's disease (AD), but studies have produced inconsistent results. We present a meta-analytic review of cross-sectional studies that examined plasma Aβ levels in AD and cognitively normal subjects, and longitudinal studies that used baseline plasma Aβ levels to predict conversion from normal cognition to AD. Medline and EMBASE databases were searched to generate an initial list of relevant studies, and selected authors approached for additional data. Twelve cross- sectional studies (n = 1483) and seven longitudinal (n = 3920) met the inclusion criteria for meta-analysis. Random effects model was used to calculate the weighted mean difference (WMD) by Review Manager Version 4.2. In longitudinal studies, cognitively normal individuals who converted to AD had higher baseline Aβ1-40 and Aβ1-42 levels (WMD: 10.29, z = 3.80, p = 0.0001 and WMD: 8.01, z = 2.76, p = 0.006, respectively), and non-significantly increased Aβ1-42/Aβ1-40 ratio (WMD: 0.03, z = 1.65, p = 0.10). In cross sectional studies, compared to cognitively normal individuals, AD patients had marginally but non-significantly lower Aβ1-42 levels (WMD:-2.84, z = 1.73, p = 0.08), but Aβ1-40 levels were not significantly different (WMD: 3.43, z = 0.40, p = 0.69). Our systematic review suggests a model of differential longitudinal changes in plasma Aβ levels in cognitively stable individuals versus those who go on to develop AD dementia. Baseline Aβ1-40 and Aβ1-42 levels in cognitively normal elderly individuals might be predictors of higher rates of progression to AD, and should be further explored as potential biomarkers.