A stochastic approach has been presented for superplastic deformation of Ti-6Al-4V alloy, and probability functions are used to model the heterogeneous phase distributions. Experimentally observed spatial correlation functions are developed, and microstructural evolutions together with superplastic deformation behavior have been investigated by means of the probability functions. The strain-rate dependent failure strain can be correctly predicted by the model. As shown by the results the probability varies approximately linearly with separation distance, and significant deformation enhanced probability changes occur during the process. Since an initial microstructure is the most crucial factor that determines the properties of final microstructure, Monte Carlo simulation has been used coupled with the probability functions for the reconstruction of microstructures. By imposing the precisely optimized distributions of phase on the test specimens, therefore finite element implementation shows better agreement with experimental data of the failure strain.