The external segment of the globus pallidus (GPe) is known as a relay nucleus in the indirect pathway of the basal ganglia (BG). In this model, the GPe receives the striatal information and projects to the subthalamic nucleus and the output nuclei of the BG, the internal segment of globus pallidus and the substantia nigra (SN). Furthermore, the pallidostriatal projections have been recently reported. In the BG, immunohistochemistry for the calcium binding protein calbindin D–28k (CB) shows the unique distribution pattern. The GPe can be divided into three (rostral and caudal CB immunopositive and medial CB immunonegative) regions according to the topographic organization of the striatopallidal projections. The aim of this study is to reveal the difference of the pallidal projection patterns originating from the three domains receiving the distinct striatopallidal afferents. We examined the distribution of axonal boutons in other BG nuclei after the micro injection of biotinylated dextran amine in each GPe subregion, combined with immunohistochemistry for CB and tyrosine hydroxylase (TH). As a result, all three GPe subregions innervated all other nuclei of the BG, although the axonal arborization of pallidal neurons mostly showed some degree of preference for the recipient domains, especially for the striatum. Interestingly, each GPe subregion projected to the SN in a different manner. The neurons in the rostral region of the GPe favored to project to the dorsal part of the SN, whereas the projection from the medial region of the GPe innervated the most part of the SN except for the rostral area. TH immunohistochemistory revealed that the rostral region of the GPe made the appositions with the dopaminergic neurons more frequently than the caudal region of the GPe did. These results suggest that distinct GPe subregions receiving the different striatopallidal afferents show the axonal trajectory in a fashion to unique itself. Since the neurons in each GPe subregions take these major courses in the different direction, the cortico–basal ganglia–thalamo–cortical loop is likely processed partially via non–parallel circuits.