This article reexamines the econometric estimation of rational-addiction mo
dels considered by Becker, Grossman, and Murphy (BGM) for cigarette consump
tion. The rational-addiction model poses a number of additional econometric
difficulties including endogeneity due to the presence of leads and lags o
f the dependent variable and serial correlation in the disturbances. BGM co
nsidered a fixed-effects two-stage least squares (2SLS) estimator. It is we
ll known that this estimator is biased for fixed T. This article suggests a
forward-filter first-difference 2SLS estimator and a generalized method of
moments type of estimator that are consistent. Using a panel dataset of 46
states over the period 1963-1992, this article estimates the rational-addi
ction model for cigarettes. Our empirical results are both supportive of th
e rational-addiction hypothesis and more plausible than BGM's original resu
lts.