In order to avoid the conventional methods requirement of selecting large quantities of input factors as well as its large errors in development of a dam seepage statistical model, problems caused by the need for many items to be considered in the earlier stage of model development, the copula entropy theory combined with partial mutual information was used to optimize the input factor selection. To obtain the copula entropy, the Gumbel function was used as the copula function, the Cauchy distribution was used to replace the normal distribution, and the Hample criterion was used to select the input factors accurately. This approach was applied to seepage detection for the Nuozhadu Dam. Comparison of the present results with those obtained from the conventional factor selection approach shows that the seepage statistical model for optimizing input factor selection based on the copula entropy has a better prediction effect.