This study analyzes the annual alcohol consumption on prices, taxes, and other economic and demographic characteristics from 50 states and the District of Columbia for 1985-2002 to study the U.S. alcohol demand behavior on beer, spirits, and wine. Young and Bielinska-Kwapisz (2003) found that the data prices contain measurement error, so using the state and federal taxes as the instrumental variables could mitigate the problem. There is the improvement in estimation to use the set of all taxes instead of each individual tax. Therefore, we use the Quadratic Almost Ideal Demand System (QUAIDS) model, proposed by Bank, Blundell, and Lewbel (1997), employing the alcohol prices or the tax instrumental variables on both pooled and clustered datasets to compare with the classical linear models. The statistical inference is based on the bootstrap variance-covariance matrix. After correct the heteroskedasticity and multicollinearity issues, the existing linear regressions models are GLS, GMM-IV, FE, RE, FE-IV, and RE-IV models. The analysis of elasticities reveals that the QUAIDS model is perform better than all linear regressions in term of explanation on consumer behavior. We use local constant and local linear estimators with the Gaussian kernel and the optimal cross-validation bandwidth to study the effects of prices and income on alcohol consumption. We also find the increasing trends of consumptions in all types of alcohol for each state. The empirical results indicate that the gross substitution effects among them could imply policy necessity on the simultaneous alcohol excise taxes imposed. In term of estimation, the instrumental tax variables could improve the QUAIDS model on reflecting the different responsiveness of price and income for light, medium and heavy consumption levels.