Objectives: Recent studies regarding the relationships between plasma amyloid-β (Aβ) levels and cognitive performance had inconsistent results. In this study, we aimed to characterize the relationship between cognitive decline and plasma Aβ levels in a largesample cognitively normal population. Methods: This population-based, prospective cohort study included 1,240 participants with normal cognition. The Mini-Mental State Examination (MMSE) was used to assess cognitive function at baseline and 2 years later. Restricted cubic splines, multivariate logistic regression, and multivariate linear regression models were used to evaluate the type of relationship between cognitive decline during the 2-year follow-up period and plasma Aβ levels (Aβ 40 , Aβ 42 , and Aβ 42/40). Results: Participants with moderate Aβ 40 levels had the highest risk of cognitive decline during a 2-year follow-up relative to individuals with low Aβ 40 [odds ratio (OR): 0.60, 95% confidence interval (CI): 0.45-0.81, p < 0.001] or high Aβ 40 (OR: 0.65, 95% CI: 0.49-0.87, p = 0.004) levels. The association between Aβ 40 and cognitive decline did not depend on sex, education level, or APOE ε4 status. There was an interaction found between age (≤ 65 and > 65 years) and Aβ 40 (p for interaction = 0.021). In individuals older than 65 years, there was a positive linear relationship between plasma Aβ 40 and cognitive decline (OR: 1.02, 95% CI: 1.00-1.04, p = 0.027). For participants ≤ 65 years old, the lower Aβ 40 and higher Aβ 40 groups had a lower risk of cognitive decline than the medium Aβ 40 group (OR: 0.69, 95% CI: 0.50-0.94, p = 0.02; OR: 0.63, 95% CI: 0.45-0.86, p = 0.004). None of relationship between plasma Aβ 42 , Aβ 42/40 and cognitive decline was found during a 2-year follow-up. Conclusion: The relationship between plasma Aβ 40 and cognitive decline was not linear, but an inverted-U shape in a cognitively normal population. The underlying mechanism requires further investigation.