The optimal working conditions for enzymes are very much elegant, and their determination is often through experimental approach, which generally is costly and time-consuming. Therefore, it is important to develop methods to use as simple as possible information to predict the optimal working condition for enzymes. Cellulase is a very important enzyme widely used in industries. In this study, we attempted to use a 20-1 feedforward backpropagation neural network to screen 24 amino acid properties related to the primary structure of cellulases as predictors to predict the pH optimum in cellulases. The results show that some predictors can predict the pH, especially amino acid distribution probability.