Since the studies of Helmholtz, it has been known that the temporal evolution of musical sounds plays an important role in our perception of timbre. The accurate temporal segmentation of musical sounds into regions with distinct characteristics is therefore of interest to researchers in the field of timbre perception as well as to those working with different forms of sound modelling and manipulation. Following recent work by Hajda (1996), Peeters (2004) and Caetano et al (2010), this paper presents a new method for the automatic segmentation of the temporal evolution of isolated musical sounds in real-time. We define attack, sustain and release segments using cues from a combination of the amplitude envelope, the spectro- temporal evolution and a measurement of the stability of the sound that is derived from the onset detection function. We conclude with an evaluation of the method.