How effective tax-favored education saving accounts (ESA) can foster the incentive of saving for child's college among heterogeneous families is still a debate. My dissertation contributes to this query by analyzing the main uncertainty of college saving. child's ability. in the context of the saving with learning model and an empirical examination of the effectiveness of ESA. This dissertation can be divided into three chapters.The first chapter develops a dynamic model combining asset accumulation and learning to explain the parents' forward-looking saving behavior when they are confronted with a real option of college choice due to uncertainty of the child's ability. The model infers, with enough time spent learning, that information can improve parents' welfare. This can be accomplished by better allocating the consumption to accommodate the burden of college cost given both asset status and child's true ability.Then, I test the implications of the model from Panel Study of Income Dynamics/Child Development Supplement \& Transition into Adulthood (PSID/CDS \& TA) (1997-2005) in the second chapter. This empirical study investigates college saving behavior when learning is present. Data suggest pessimistic and/or rich parents might reduce the college saving, which confirms the interaction of wealth and learning effects predicted by this model. The result also supports the state dependence of parents' college expectation and diminishing persistence over time due to learning.The third chapter is devoted to a policy study on education saving accounts (ESAs). I examine whether the contribution to ESA is associated with education saving propensity (ESP) among families with different income by using Survey of Consumer Finance (1998-2007). The model is estimated by a dynamic synthetic panel. The evidence suggests that only the poor families significantly utilize ESAs to fulfill ESP. Relatively richer families might mainly treat ESA as a tax shelter. I then provide the justification on the ineffectiveness of ESA by using my theoretical model. A number of fiscal policy improvements are proposed to encourage early learning child's ability in an effort to raise the economic efficiency of ESA.