Abstract.Composite structural members experience extensive and complex damage that accumulate in a relatively steady pace as the structure is quasi-statically loaded. This damage progression which starts as matrix cracks, delaminations, and random fiber breaks, turns unstable when groups of adjacent fibers, ranging from four to ten fibers fail together, after about 85% of ultimate strength, as reported in the literature. Identifying this critical damage that precedes the final fracture has been difficult even in laboratory specimens. There is little consensus on successful use of AE signals to differentiate failure modes. The inability of AE patterns to identify failure modes is likely caused by the limited frequency bandwidth of available AE sensors, and the high attenuation seen in AE signals particularly in the frequency range likely to be associated with fiber fractures. As a part of this study new acoustic emission sensors capable of measuring frequencies to 2 MHz were developed. In addition, composite specimens were instrumented with sufficient number of sensors to capture high frequency signals before they are attenuated. Unidirectional, cross-ply, and quasi-isotropic carbon-epoxy composite tensile specimens were monitored while they were statically loaded to failure. Distinctly different signals corresponding to the three failure modes could be observed. High frequency acoustic emission signals with frequencies well in excess of 1MHz, mostly seen in the last 20% of the loading cycle. Signals with frequencies in the range of 300 kHz to 700 kHz and duration of the order of 50 microseconds, were observed in cross ply and quasi-isotropic specimens, and are believed to be from matrix cracks. Fewer events with frequencies below 300 kHz and duration that exceeded about 200 microseconds are believed to be from delaminations. An important observation in this study is the appearance of groups of near identical waveforms, which are believed to be from clusters of adjacent fiber breaks, appearing in the last 15% of the loading cycle. The size of the individual groups of such AE signals and their number of groups increase as the final failure is approached. Hence, by monitoring such groups of AE waveforms, it may be possible to recognize the point at which the damage growth is turning unstable and it may still be possible to avert catastrophic failure.