The objective of this study was to compare different methods of estimating the demand equations of global forest sector models. Past applications have shown that model projections depend critically on the elasticity of demand with respect to income and prices. To measure the elasticity coefficients, nine groups of products were examined, ranging from fuelwood to paper and paperboard, in 64 countries. The analysis was done with panel data on consumption, price and income, from 1973 to 1997. Two alternative forms were hypothesized: a static demand model, and a dynamic partial adjustment model. The demand equations were estimated first with classical panel methods: ordinary least squares pooling of country data, least squares with country dummy variables, between group pooling and error-component models. However, rejection of the hypothesis of homogeneity of elasticities across countries led to the investigation of two shrinkage estimators: the Stein rule and the iterative empirical Bayes estimator. Within-sample predictions were carried out to compare the accuracy of each estimator. The Stein rule estimator, based on the static model, gave the smallest root mean square error. Nevertheless, because the Stein-rule estimates often had theoretically wrong signs and large variances, other estimators, although biased, were deemed preferable for policy analysis. In conclusion, pooling by ordinary least squares with country dummy variables was one of the most useful methods to obtain demand equations applicable to many countries. Furthermore, static models predicted demand better than dynamic models, conditional on price and national income.