Numerous studies have shown a marked decrease of beta-amyloid(42) (Abeta(42)) in the cerebrospinal fluid (CSF) of patients with incipient Alzheimer's disease (AD). However, studies on Abeta in plasma are contradictory, and show very marginal differences between patients and controls. Here, we analyzed plasma samples using a new multiplex immunoassay for simultaneous analysis of Abeta(1-40), Abeta(n-40), Abeta(1-42), and Abeta(n-42). The plasma samples were obtained at baseline from two independent cohorts of patients with mild cognitive impairment (MCI) and age-matched controls. In the first cohort, 41% of the 117 MCI cases converted to AD during a clinical follow-up period of 4-7 years. In the second cohort, 14% of the 110 MCI subjects developed AD during a clinical follow-up period of 2-4 years. None of the plasma Abeta isoforms differed between MCI patients that subsequently developed AD and healthy controls or stable MCI patients. The Cox proportional hazards model did not reveal any differences in the probability of progression from MCI to AD related to plasma Abeta levels. In contrast, low levels of Abeta(1-42) in CSF were strongly associated with increased risk of future AD. The absence of a change in plasma Abeta in incipient AD, despite the marked change in CSF, may be explained by the lack of a correlation between the levels of Abeta(1-42) in CSF and plasma. In conclusion, the results show that CSF biomarkers are better predictors of progression to AD than plasma Abeta isoforms.