Long-term stability assessment of slope characterized by the presence of red clays depends essentially on the creep model of red clay adopted. A specially designed device was used to conduct the tri-axial creep tests on red clay specimens. Deviatoric stresses were imposed on the specimens by stepwise loading，which were consolidated under varying confining stresses. The full-process creep curve of axial strains with increasing deviatoric stress was transformed equivalently to a cluster of creep curves under each stress levels by using "Chen's method". Furthermore，the ultimate deviatoric stress of red clay specimens before yielding in creep tests under varying confining stresses was determined using isochronal curve method，and used to establish the long-term shear strength criterion. A creep model of red clay in terms of axial strain accounting for the deviatoric stress and confining stress was established in the framework of ANFIS neural network. In this framework，a part of test data were used in training the creep model，while the remaining test data were used to examine its capability of predicting the creep response of red clay. A good agreement between the measurements and predictions validates the effectiveness and accuracy of this presented creep model.