Abstract: ITS, trnL-F and trnS-G regions of 20 species of Caragana in China were sequenced, and these sam-pled species covered all recognized subgenera and sections in this genus. Separate phylogenetic analyses of the above DNA fragments generated concordant topologies and a combined analysis of these three datasets signifi-cantly improved the statistical supports of all identified clades. Species of subgen. Caragana firstly diverged as a monophyletic group in the phylogenic tree, which was congruent with morphological study and geographical distribution of this taxon. Representative species of ser. Occidentales and ser. Bracteolatae formed one mono-phyletic clade, indicating that ser. Occidentales should be placed in sect. Longspina, which contains ser. Bracteo-latae, but not in sect. Spinosae or sect. Pruinosa. Previous conclusion based on morphology that two series should be defined in sect. Frutescentes according to the shape of leaves was well supported by molecular phylogenetic data in the present study. Because the species of ser. Dasyphyllae had close relationships with those of ser. Spino-sae, phylogenetic analysis did not merit the definition of two separate series in sect. Spinosae. Species of Subgen. Jubatae clustered together as a monophyletic clade, and thus should be put in one section, sect. Jubatae. In sect. Jubatae, species having legumina with smooth interior and those producing legumina with haired interior were respectively nested within two separate subclades, corresponding to two series in this section. In the phylogeny tree, sect. Tragacanthoides, defined by Sanczir, was shown to be polyphyletic, suggesting that species of sect. Longspina, sect. Spinosae and sect. Jubatae should be transferred out of this section. Caragana ordosica and C. tibetica had been treated as one species, yet in this study they were proved to be two different species, by genetic distance based on ITS sequences.