In this work, an evolutionary neural network (ENN) is proposed for forecasting net flow of a car sharing system. This work consists mainly of two contributions. The first is to develop a mixed optimization approach with genetic algorithm (GA) and back propagation (BP) training for the ENN. In particular, the crossover operator of the genetic algorithm is performed with multiple neural networks that have heterogeneous structures: either different number of nodes in a hidden layer or different number of hidden layers. Hence, this optimization process enables co-evolution of multiple NN structures which present different nonlinear models, and facilitates the selection of the most suitable forecasting model from multiple candidates. To expedite the searching process for ENN and meanwhile retain an efficient learning rate, the back- propagation training is applied only to the best or the second best chromosome in each generation. The second contribution of this work is the application of the ENN to a real forecasting problem arising from a car-sharing system. Despite the presence of randomness, nonlinearity and complexity in the forecasting process, the ENN demonstrates superior performance when comparing with both classics time series forecasting approaches and other soft-computing approaches.