Abstract: The design of randomized blocks with the few-tree plots has been widely adopted in experiments in tree breeding, since it is more efficient in utilizing land, and can ensure completeness of experimental data. In this design, each plot is a smallest statistical unit, and the mean of each plot can only be obtained by measuring all the trees in a plot or by sampling. However, since this work needs a lot of time and money, it is desirable to measure the best tree in each plot and adjust the result instead.In a study of the provenance test of Chinese fir (see pp35-45, NO. 4 , 1980, Journal of NTCFP), the authors developed a method of measuring the best tree in each plot, in the design of randomized blocks with the few-tree plots in the statistical analysis, the results of which can be summarized as follows:1. During the four years of investigation the order of the means of eleven provenances of Chinese fir measured for the best tree in each plot was similar to the order of the mean measured for all trees in each plot, the correlation coefficients of the order being 0.9807 (1976) , 0.9913 (1977) , 0.9909 (1978) and 0.9947 (1979) .2 . There existed little fluctuation in the order of the best tree in each plot in the period of study. The position of a specific provenance might drop from the first place to the second in one year, and in another year might come up to the first again. But it has not been found that a best tree dropped from the first place to the third or to the fourth.3 . Difference between the expected mean squares arrived at by using the two different statistical units tended to decrease year by year.4 . With some adjustment, the heritability and genetic gains of the selected characteristics of these provenances worked out by using the best tree as unit of statistics can as well be used to draw correct conclusions.