Classification system of flow regimes in a vertical pneumatic conveyor is investigated by using a Time-Delay Neural Network (TDNN), which is a multilayer feed-forward network originally designed for speech recognition. Time series spectrum of electrostatic charge sensor is used to characterize flows. The recognition performance of the network after the training are excellent except for a boundary regime. A corresponding multi-layer perceptron (MLP) shows almost comparable recognition performance but TDNN shows higher generalization ability than MLP. The resulted automatic classification network of flow regimes will easily extend to an on-line qualitative interpreter and will be integrated with flow rate estimation algorithms to improve the accuracy and reliability of the sensor system.