Three methods of estimating confidence intervals for the parameters of Weibull nonlinear models are examined. These methods are based on linear approximation theory (Wald), the likelihood ratio test, and Clarke's (1987) procedures. Analyses are based on Weibull dose-response equations, developed from the National Crop Loss Assessment Network (NCLAN), that estimate yield as a function of ozone concentration. Comparisons among the three methods of confidence interval construction were also made for relative yield loss, a nonlinear function of the Weibull parameters. The results of these comparisons are considered along with the conclusions indicated using Clarke's measures of parameter-effects curvature and his seriousness criteria. Plots of the Wald and likelihood ration confidence intervals are shown for comparison. It is shown that in two cases the Wald confidence intervals are misleading, but in a third case they are entirely adequate. Clarke's methods identified the two cases where the linear approximation is inadequate and also showed whether his adjustment to the Wald would result in acceptable confidence intervals. The failure of the linear approximation appeared to be due to high variability and/or incomplete coverage of the response curve by the data. Comparisons of the Wald and likelihood ration confidence interval estimates for six other data sets showed the linear approximation to be adequate.