The invention discloses a double-cluster geometric channel modeling method based on a three-dimensional space. The method comprises the following steps: firstly, building a large-scale MIMO V2V channel model based on a geometric relation in a confocal semi-ellipsoid scattering scene; secondly, using the concept of clusters to describe scatterers, wherein many scatterers exist in the clusters, andthe clusters are randomly distributed on the semi-ellipsoidal surface; supposing that the number of the scatterers in the clusters approaches infinity, and respectively expressing the continuous random variables alphaT<nl,1>, betaT<nl,1>, alphaR<nl,2> and betaR<nl,2> by discrete variables alphaT<l,1>, betaT<l,1>, alphaR<l,2> and betaR<l,2>; then, introducing a birth and death algorithm to simulatedisappearance and evolution of the clusters on a time axis and an array axis; and finally, deducing the statistical characteristics of the proposed channel model according to the complex channel impulse response. The performance of the communication system of the V2V in the semi-ellipsoidal environment is improved by calculating the geometric statistical characteristics of the three-dimensional double-cluster semi-ellipsoidal body.