The modelling of tendon behaviour during failure stages is nonlinear and heavily random. However, the understanding of its behavior during such stages, and development of models that can give an accurate prediction of its behavior during failure can provide a means for developing effective tendon therapies. This study is aimed at demonstrating the capability of an artificial neural network in the modelling of failure phases in tendons. A nonlinear autoregressive with exogenous inputs network is applied to three different tensile test data of the human supraspinatus tendons. Owing to data scarcity, the network was trained using two different test data which were randomly sampled and divided into 50%, 25% and 25% proportions for training, validation and preliminary testing. The third test data were used for the final testing phase. The procedure was cyclically performed for each of the results that have been presented in this study. The neural network predictions are presented as curves fitted over actual test results with corresponding error plots. The results indicate that the network is able to accurately predict the failure behaviour of these tendons with correlations of above 99 % for all tests. This is an excellent and very promising result in the light of the difficulties that most deterministic mechanistic models encounter in the modelling of soft tissue failure behaviour. With further development of this technique, sports and exercise physicians would enhance knowledge in mechanisms of tendon failure and be able to devise more injury preventive strategies.