Abstract: Two一years experiment data using 40 parents of upland cotton (G, hir- sutum L,)with 14 measured characters demonstrated that the phenotypic mean correlation (RP) matrix can be used to replace the Re matrix as the foundation of the distance estimation of the paraents, The former had a good stablity in two years, which character standardizing was rational and was always positive definite matrix, The character can be used in the ana- lysis of phenotypic distance (DPZ)absolutely sameness,During two years，using the Rp had higher similarity of DPZ and cluster results than the Rg, Compared to the different methods of cluster analysis, the longest distance method had a larger range of the cluster, higher sensitivity, rational result of the clus- ter and higher using value on the cotton breeding, The authers suggested that using phenotypic correlation matrix (P)一principal components transformation (C)-Euclidean distance(E)-the longest distance method of clustering(L), i, e, PCEL method be the best way for the distance estimation and cluster analysis of the cotton paraents,

Key words: Cotton, Correlation matrix, Distance estimation, Cluster analysis