A study of the mechanical properties of an unsaturated residual slope soil is of the engineering significance. General laboratory triaxial tests are carried out to obtain the stress-strain relationship between the unsaturated residual slope soil and the changes in water content. Improved BP neural network prediction model is established based on test data. The results from comparing predicted values and measured values show that this network prediction model has good fitting precision and good generalization ability. The results can also provide a reference for engineering applications.