We fit the Florida Model with an AR(1) error structure to pooled cross-country International Comparison Project (ICP) data of Seale, Walker, and Kim and estimate the model with the minimum information (MI) estimator. Point estimates obtained by MI are similar in value to those obtained by Seale, Walker, and Kim with maximum likelihood (ML). Two similar simulations but with different sample sizes are conducted to compare the relative efficiencies of MI and ML with known and unknown (MLU) covariances. In the larger sample, the MLU is more efficient in terms of root-mean-squared errors (RMSEs) than the MI. Noteworthy, in the small sample, the MI is more efficient in terms of RMSEs than MLU, even though MLU explicitly accounts for AR(1), whereas the MI does not. These results correspond to earlier findings of Theil for time-series and cross-sectional data.