genotypes of cotton
germplasm were used as an example for comparing the efficiency of establishing core collection based
on five quantitative fiber traits, which were the 2.5% length (mm), uniformity (%), strength (gf/tex), elongation (%), micronaire. Genotypic values were predicted unbiasedly by AUP (Adjusted Unbiased Prediction) of mixed linear model. Based on the genotypic values, core collections were constructed at 30% sampling proportion by using two genetic distances (Mahalanobis
distance, Euclidean distance), seven hierarchical cluster methods (Single linkage, Complete linkage, Median method, Centroid method, Unweighted pair-group method using arithmetic averages (UPGMA), Weighted pair-group average (WPGMA), Ward's method and three sampling strategies (random sampling, preferred sampling, deviation sampling), respectively. The genetic variation of quantitative traits among core collections was compared by evaluating the means, variances, ranges and coefficients of variation of the traits. The results showed that the Mahalanobis distance was much better than Euclidean distance in
constructing core collection; generally, the preferred sampling strategy was better than the deviation sampling, although the deviation could also significantly increase the variance and coefficient of variation; among the seven hierarchical cluster methods, single linkage was the best linkage rules for constructing core collection, which could capture the most genetic diversity of quantitative traits, the succeeding methods were the median method, centroid method and UPGMA.