Bits currently are selected on the basis of performance of similar bits in offset wells. The relationship between formation properties, drilling-fluid characteristics, bit design, and operational parameters is not easily understood. A three-layer artificial neural network was designed and trained with held data. Results indicate that the back propagation architecture with two hidden slabs is the most effective neural-network design for predicting optimum bit type. The new model successfully predicted bit types for several fields. Correlation coefficients for the predicted and field-used bit types ranged from 0.857 to 0.975.