This research had the objectives of clustering, the most similar seeds matrixes by species, using the observed variables in analysis of seed samples of native forest species in the Center of Forest Researches and Conservation of the Soil, Santa Maria, Rio Grande do Sul, since 1997 up to March 2001. For the cluster analysis, all the species with four or more seed analysis by analyzed in collected lot, by the hierarchical complete method of the standardized euclidian medium distance as measured of similarity. The dendogram was constructed using the method of complete linkage. The technique of principal components for the reduction of the number of variables was also used. The gender Schinus sp. and the native species Cassia leptophylla Vogel., Peltophorum dubium (Spreng.) Taub., Cedrela fissilis Vell., Allophylus edulis (A. St.-Hil., Cambess. & A. Juss.) Radlk., Lafoensia pacari A. St. Hil., Enterolobium contortisiliquum (Vell.) Morong. e Apuleia leiocarpa (Vogel) J. F. Macbr., were the species in which the third and fourth principal components explained 80% the existent variation of the data with the humidity and percentage of abnormal germination seedlings eliminated in the majority species and the percentage of normal germination seedlings and dead seeds found in all the groups . The cluster analysis was efficient in the separation of the groups of all the tested species and origin influenced a little in the formation.
Euclidian distance; principal components; cluster analysis
