Abstract Two approaches toward fitting regression models with multiplicative error heteroscedasticity recur in the forestry, ecology, and statistical literature. One includes the estimation of the heterogeneity in the fitting process. The alternative approach entails the use of variance estimators that are robust to the error variance heterogeneity. Under suitable conditions, the former method offers nonnegligible gains in efficiency, whereas the robust alternatives provide accurate assessment of ordinary least squares estimators even in the presence of heteroscedasticity. The performance of both approaches are examined and contrasted, and suggestions for future applications and research are made on the basis of these results. For. Sci. 35(1):105-125.