In order to improve the integer ambiguity resolution successrate and the baseline precision, a strategy of stochastic modeling for precise relative positioning aimed at hybrid constellation BDS was proposed, namely the hybrid stochastic modeling strategy. In the new strategy, variances of singledifference observations were estimated with the leastsquares variance component estimation. After that, the observation variances of different orbits satellites were fitted with different models, carriertonoise ratios dependent model was used to fit the observation variances of geostationary earth orbit satellites, and two individual elevationdependent models were used to fit observation variances of inclined geosynchronous satellite orbit and medium earth orbit satellites. Then the stochastic model of observation model was constructed in real time with these different fitted models. The experimental results reveal that, when compared with the traditional, simplified model and the sole elevation or carriertonoise ratios model, the proposed hybrid stochastic model is more realistic in reflecting the noise property of observations of different orbits satellites. And it has an overall better performance in integer ambiguity resolution success rates and relative positioning precision. Hence, the hybrid stochastic model is more suitable for BDS in high precision positioning applications.