Proper estimation of input parameters plays a crucial role in the numerical simulations of geotechnical engineering. In conventional numerical procedures, all the input parameters are assumed constant and unique, and their variability and cross-correlation are usually neglected, so that significant error can be induced in the simulated results. To resolve the problem, an inversion method must be introduced to obtain mechanical parameters, which are spatially correlated. The random field of a single parameter has a characteristic of auto-correlation, whereas the random field of multi-parameter is not only auto-correlated, but also cross-correlated. Based on ‘two sides of one’ (auto-correlation ??cross-correlation ??remaining auto-correlation), a mathematical inversion model of multi-parameter random field is built. Correlation statistics results of random fields generated by ‘two sides of one’ method show that auto-correlation and cross-correlation are well satisfied in the isotropic correlated condition.