The "chunk' phenomenon has attracted some attention lately. In the auditory domain a chunk is seen as a segment of sound. This work first investigates theories of chunks from cognitive science and musicology. We define the chunk as a closed entity of a certain size and with a dynamic structure containing a peak. The perceptual analysis of three songs leads to a collection of chunks, classified according to their internal structure. Common signal processing methods are then used to extract loudness, pitch and brightness of the chunks. The features are modeled in a simple chunk model, consisting of height, slope, peak and peak position, and the values of the model are furthermore estimated for each chunk. Histogram plots are used to give values for the probability of each chunk type. The model parameters from the features are finally compared to the parameters found by listening and determining the shape of each chunk.