The cause and progression of a crack in operation inmportant hydraulic concrete structures are affected by hydrostatic pressure and temperature. Different combinations have different influences on the crack. A new data mining approach to the unfavorable load combinations of the crack is presented using adaptive - network based on fuzzy inference system (ANFIS). Compared with the method to extract unfavorable load combinations by experience and measured data, this method can avoid the impact of subject. The calculated results obtained from some engineering illustration show that the conclusion obtained by this method is in good agreement with observed data. The rule of the unfavorable load combination of the crack is generated.