Attempt is made to undertake comparative study of feedforward neural network and Logit regression model, whereby some theoretical results and properties are clarified with their application to practical problems. Also, some aspects of approximation, overfitting and model selection are examined analytically and numerically.