Most of the existing channel model for multiple-input multiple-output (MIMO) vehicle-to-vehicle (V2V) communications only considered that the terminals were equipped with linear antenna arrays and moved with fixed velocities. Nevertheless, under the realistic environment, those models are not practical since the velocities and trajectories of mobile transmitter (MT) and mobile receiver (MR) could be time-variant and unpredictable due to the complex traffic conditions. This paper develops a general 3D nonstationary V2V channel model, which is based on the traditional geometry-based stochastic models (GBSMs) and the twin-cluster approach. In contrast to the traditional models, this new model is characterized by 3D scattering environments, 3D antenna arrays, and 3D arbitrary trajectories of both terminals and scatterers. The calculating methods of channel parameters are also provided. In addition, the statistical properties, i.e., spatial-temporal correlation function (STCF) and Doppler power spectrum density (DPSD), are derived in detail. Simulation results have demonstrated that the output statistical properties of the proposed model agree well with the theoretical and measured results, which verifies the effectiveness of theoretical derivations and channel model as well.