We investigate the robustness of various estimators of the mean for two families of symmetric distributions (exponential power and t) indexed by the kurtosis γ and Hogg's (1972) measure of tail thickness Q. For fixed γ or Q, the optimal estimator for one family is often inefficient for the other family. Furthermore, over various ranges of γ or Q some common estimators (e.g., the median) are efficient only for one family. However, other estimators (e.g., some trimmed means and Gastwirth's three-percentile estimator (1966)) do maintain good efficiency over a wide range of γ or Q.