Prior prediction of burst pressure of the composite pressure vessels well before its failure would be a complimentary method in the area of composite characterization. In this proposed research, an attempt was made to predict the failure pressure of  the composite pressure vessels. A series of five identical GFRP (Glass Fiber Reinforced Plastics) pressure vessels were monitored with an acoustic emission (AE) system, while proof testing them up to 50% of their theoretical burst pressure. Back propagation neural network models were generated for the prior prediction of burst pressure of the composite pressure vessels. Three different networks were developed with the peak amplitude distribution data of acoustic emission collected up to 30%, 40% and 50% of the theoretical burst pressures. Amplitude frequencies of AE data recorded from three bottles in the training set and their corresponding burst pressures were used to train the networks. Only the amplitude frequencies of the remaining two bottles were given as input to get the output burst pressures from the trained networks. The neurons present in the multi-hidden layers of the networks were able to map the patterns of failure present in the AE data. The results of three independent networks were compared, and it was found that the network trained with more AE data had better prediction performance. Prior prediction of burst pressures of the composite pressure vessels at low proof testing level may serve to avoid significant fiber failures and the associated structural integrity degradation.