needs for forestry herbicide trials. Can. J. For. Res. 23: 2153Res. 23: -2158 Forest herbicide experiments are increasingly being designed to evaluate smaller treatment differences when comparing existing effective treatments, tank mix ratios, surfactants, and new low-rate products. The ability to detect small differences in efficacy is dependent upon the relationship among sample size. type I and II error probabilities, and the coefficients of variation of the efficacy data The common sources of variation in efficacy measurements and design considerations for controlling variation ate reviewed, while current shortcomings are clarified. A summary of selected trials estimates that coefficients of variation often range between 25 and lOO%, making the number of observations necessary to detect small differences very large, especially when the power of the test (1 -l3) is considered. Very often the power of the test has been ignored when designing experiments because of the difficulty in calculating f3. An available program for microcomputers is introduced that allows researchers to examine relationships among sample size, effect size, and coefftcients of variation for specified designs, a and p. This program should aid investigators in planning studies that optimize experimental power to detect anticipated effect sizes within resource constraints. ZEDAKER, S.M., GRE~~IFUZ, T.G., et MILER, J.H. 1993. Sample-size needs for forestry herbicide trials. Can. J. For. Res. 23 : 2153-2158. Les experiences sur les herbicides forestiers. sent de plus en plus structtmfes pour evaluer de petites diff&ences entne les traitements lorsqu'on compare les traitements actuellement utilis& les rapports des m&anges dam les reservoirs, les surfactants et les nouveaux produits utilis& 8 faible dose. La cap&u5 de d&ecter de p&es differences d'effkaciti d&pend de la relation entre la dimension des &iantillons, des eneurs en probabik? de types I et II et des coeffkients de variation des doM6es d'effrcacid. Les sources communes de variation dans les mesurea d'efIicacit6, et les consid&ations de structure pour contiler la vatiatio~ sont revis& pendant que les dkfauts courants sont clarifi&~ A l'aide d'un sununaim d'l%&S s&ctio~&s, on estime que les coefficients de variation se situent souvent eatre 25 et 100%. ce qui n&es&e un t&s grand nombre d'observations pour d&e&r de pet&es differences surtout lorsque la puissance du test (1 -B) est consid&&. Tti?s souvent la puissance du test, a &? &nor& dans la structure des experiences a cause de la difficuld & cakuler p. Nous inuoduisens un prograuuue disponible sur mko-oniinateur qui permet aux chercheurs d'examiner lcs relations entre la' dimension des kchaatillons, les effets de la dimension et des coefficients de variition pour des structures do~&s,