Due to the excellent energy-saving and environmental protection features, electric vehicles (EVs) are gaining significant market penetration, especially in densely populated urban areas with systemic air quality problems. What follows is that we will face one of the biggest challenges: how to accurately predict the energy consumption (EC) of electric vehicles to alleviate the ‘range anxiety’problem. To address this problem, this paper proposes a hybrid approach using short-trip segment division (S-TSD) algorithm and deep neural network (DNN) for predicting the energy consumption of battery electric vehicles (BEVs) by using real-world driving datasets. In this study, we aim to accurately predict energy consumption at the short-trip level, so we propose a novel S-TSD algorithm to divide the driving process of BEVs into several driving segments and conduct feature extraction for each segment. Then, using the extracted features, we develop a series of DNN models with varying hidden layers to investigate the respective performance for EC predictions. The results show that the optimal number of hidden layers is identified as 5, and the neuron structure for each hidden layer is determined as 13× 256, 256× 512, 512× 256, 256× 128, and 128× 1, respectively. Two state-of-art models are selected as benchmark methods to compare with the proposed model for the EC prediction tasks. The comparative analyses suggest that the proposed model outperforms the benchmark methods in terms of the four index values of MAE, MAPE, MSE, and RMSE, indicating that the proposed model has a good performance for predicting the energy consumption of BEVs at a short-trip level.