A lot of studies have been done on Acoustic Emission (AE) covering a wide range of materials and applications. Within a structure that is under loading, AE has proven to be useful in determining the type, location and accumulation of damage within a material. The general goal is to use AE for inservice monitoring of structurally loaded parts. However, more research is necessary before the method can be employed in real-time applications. The objective of this project is to contribute to the development of algorithms that classify failure modes real time in composites using AE. Composite material specimens are loaded under tension while recording AE, after which the signals that are recorded are used to create an independent damage mode signature. This leads to the ability to classify damage modes per time increment. Mechanical tests are performed on specimens while recording AE signals. Carbon fiber reinforced polymer specimen with unidirectional 90∘layup are loaded in tension up to failure to create a fingerprint of the matrix cracking damage mechanism. The same material with a crossply layup is loaded in tension under quasi static and fatigue loading. The fingerprint is then used to separate signals coming from this type of damage and other damage mechanisms. Next to the basic parameters of these signals, additional features are generated by processing the waveform of each signal. The wavelet transform is used to find frequency related features that characterize each signal. Machine learning algorithms are developed and used to cluster and classify the AE Signals. Overall, the real-time clustering algorithms have proven to be successful in clustering and classifying incoming AE signals in an efficient way. The system was able to correlate incoming data to matrix cracking data, within a very limited time frame. With room for optimization the proposed methods seem fit to be applied in real-time applications. Before this can be done however, more testing and validation of these methods is required. The most valuable addition to this study would be a proper validation of the clustering results. Regarding improvement of the system, the main bottleneck of the proposed algorithms is loading large datafiles. This can be solved by reading data in batches, but requires further development of the algorithms.