Leaves of nine kinds of plants from three subfamily of Rosaceae were used as materials. Genetic relationship was analyzed and species were identified through studying FTIR of nine kinds of plants. Leaves mainly contain large amounts of carbohydrates, proteins, lipids, nucleic acids and other substances. The peaks of carbohydrates are mainly between 1440 and 775 cm(-1). The vibration peaks of the cellulose and lignin are between 1440 and 1337 cm(-1). The peaks between 1000 and 775 cm(-1) are stretching vibration of ribose. The vibration peaks of protein are between 1620 and 1235 cm(-1). The peak at 1620 cm(-1) is sensitive to C=O stretching vibration of protein amide I. The peak at 1523 cm(-1) is assigned to N-H and C-N stretching vibration of protein amide II. Peaks of lipids mainly appeared between 2930 and 1380 cm(-1). The peak at 2922 cm(-1) is CH2 stretching vibration of fat. The peak at 1732 cm(-1) is C=O stretching vibration of fatty acids. The mark peak of the nucleic acid appears in the region between 1250 and 1000 cm(-1). The peak at 1068 cm(-1) is due to the symmetric stretching vibration of PO(2-) group of the phosphodiester-deoxyribose backbone, and the peak at 1246 cm(-1) is associated to the asymmetric stretch vibration of PO(2-) group. The results showed that the cluster model is established by smoothing, standardizing, the second derivative, principal component analysis and Hierarchical cluster analysis. It is accordant with the traditional classification. The result of cluster shows that Prunus armeniaca L. and Prunus seudocerasus Lindl. were clustered into one (Prunoideae). Potentilla fulgens Wall. Rosa chinensis Jacd and Fragaria ananassa Duchesne var. were clustered into the second (Rosoideae). Pyracantha fortuneana Li, Malus pumila Mill. Eriobotrya bengalensis Hook. f. and Malus hallianna Koehne were clustered into the third (Pomoideae). The correct rate of cluster at subfamily is 100%. The correct rate of cluster at genus is 55.56%. The correct rate of identification is 100% when unknown species waiting for determined were laid into the model of Hierarchical cluster to identify. This study provides a new thought and method for genetic relationship analysis of planst.