Abstract: Determination of soil probability model parameters is the foundation of reliability design in geotechnical engineering. In order to promote the reliability design of geotechnical engineering, regional probability model of soil index should be established firstly. As a reflection of autocorrelation characteristic of soil index-correlation distance, it's necessary to study its regional distribution characteristics and establish the probability distribution model. This paper collects 176 drilling CPT data of Xi'an loess. It uses the resistance value qc as sample and calculates the correlation distance of Xi'an loess by spatial recursive method. It analyzes the statistical characteristics, estimates the unilateral confidence interval of average value. Representative value of Xi'an loess correlation distance in each layer is put forward. Probabilistic model of Xi'an loess correlation distance is established. The model parameters are estimated. The goodness- of -fit test on the model is conducted using the Pearson chi-square method. The test results show that the model complies with the Beta distribution.