The performance of a number of estimators--simple expansion, ratio, unbiased ratio, regression, and unequal probability types and stratified sampling with the simple expansion estimator--was examined in three forest tree populations using sample sizes of 4, 12, 24, and 40. Estimates were constructed for total volume, height, and crown area for these test populations. Independent variables employed were diameter, height, and crown area plus several transformations and combinations of these terms. Relative performance was evaluated using estimates of sampling variances and biases obtained from repeated sampling of the test populations. For the larger sample sizes studied, linear and parabolic regression and the Horvitz-Thompson pps estimator were usually among the best three estimators. For the smaller sample sizes, linear regression, Horvitz-Thompson pps, and ratio-of-means estimators were best. Population characteristics affecting estimator performance and implications for practical problems are discussed. Forest Sci. 17:2-13.