Up to now, tunnel cutting face evaluation problems have tended to incorporate the engineer's subjective views, and have not agreed with empirical knowledge. As utilizing subjective views as well as empirical data is also effective from the viewpoint of engineering, we studied mechanisms for making tunnel cutting face evaluations using the professional engineer's empirical data. Since tunnel cutting face evaluation problems are simply pattern recognition problems, we decided to utilize neural network technology by artificially reproducing empirical data on computers. Using this technology, the back-propagation method is commonly used for pattern recognition. However, this method has some defects, such as falling into a local minimum. In this study, we traced back to the perceptron, which no longer attracts attention, and found that the defects of the back-propagation method can be solved by modifying it. We then applied this modified perceptron to tunnel cutting face evaluation problems and concluded that this method is useful.