This thesis proposes a hypothesis that managerial overreaction of past earnings performance which can affect managers’ dividend decisions. This proposition claims that managers can extrapolate earnings growth/decline over the past few years into the future, and then managers may make dividend decisions based on their expectation of such extrapolation. Managerial extrapolation is different from managerial overconfidence, although they both belong to managerial overreaction. To examine this hypothesis, we follow Fama and French (2001), Baker and Wurgler (2004a) and Hoberg and Prabhala (2009)’s methodology to test whether managerial extrapolation which is represented by firms’ earnings persistent growth/decline over the past five or three years can affect managers’ decisions to initiate/continue/increase/pay/omit/decrease dividends, given other conventional determinants including size, investment opportunities, profitability, and risks. We find robust evidence that indicators of managerial extrapolation have significant effects on firms’ dividend decisions during the in-sample period from 1963 to 2000, the out-of-sample period from 2001 to 2013, and the whole sample period from 1963 to 2013 in the U.S. market before or after controlling for recession or financial crisis. Further, we find that firms who initiate/continue/increase/pay dividends after experiencing past growing earnings do not show future growth in profitability, while firms who omit/decrease dividends after experiencing past declining earnings do not show future decline in profitability. Our findings hold when we use alternative measurements to form proxies of managerial extrapolation by using positive EPS only, by using past sales growth/decline, or by using past free cash flow growth/decline. Importantly, we find that the significant impact of managerial extrapolation on firms’ dividend decisions is not affected after considering other managerial behaviour factors including catering incentive to dividends (Baker and Wurgler, 2004a) and different measurements of managerial overconfidence based on option-holding activities, investment ratio, or managers’ net-buying activities on their own firms’ shares.