Identifying simple tree attributes of street trees, i.e., tree height, crown width and crown height obtained by unmanned aerial vehicles (UAV), plays a significant role in urban management to maximize the ecological benefits of street trees. However, simple attributes usually fluctuate over a wide range due to differences in tree-age and growing environment, leading to inconspicuous interspecific features and low classification accuracy. Composite attributes, expressed by two or more simple attributes, can be used to reduce the variability in simple tree attributes, thus providing an alternative to improve the accuracy of street tree classification. In this study, we examined the classification effects of simple attributes and simple-composite attribute combinations by back propagation (BP) neural network, K-nearest neighbor (KNN) and support vector machine (SVM). The results showed that (1) the values obtained by UAVs and observations were highly consistent and R2 values for tree height, east-west crown width, north-south crown width and crown height were 0.90, 0.87, 0.78 and 0.76, respectively. The relative errors of tree height were the most stable among different tree species, followed by the crown height, east-west crown width and north-south crown width. (2) Compared to simple attributes, composite attributes displayed significant differences among street tree species, and these differences were helpful for identifying street tree species that could not be identified with simple attributes. (3) The accuracy of tree species identification after including corresponding composite attributes can be improved by 29.7% (kappa coefficient improved by 0.34) compared with only using simple attributes. The results suggested that consideration of composite attributes in street tree species classification reduced the mistakes for identifying tree species, thus providing a new approach for identifying street tree species and managing street trees efficiently.