A multiple input, two output model is proposed which enables the two microphone acoustic intensity method to decompose the intensity vector into contributions from individual sources, even when they are coupled and in close proximity within the measurement surface. By treating characteristic signals from each source as the inputs, and the sound pressure signals from the two closely spaced microphones as the outputs, the model's frequency response functions are developed from a least squares approximation. The cross spectrum between the two microphones can then be expressed as a function of the input signal spectra and the model's frequency response functions. By manipulating the model terms the selective cross spectrum associated with the radiation from each individual source can then be estimated. The selective cross spectrum is then processed via standard methods to obtain the acoustic intensity vector from each source. A series of laboratory experiments is summarized which demonstrates that the technique can accurately decompose the acoustic intensity vector from highly coherent sources ( γ122 > 0·9) buried in background noise in a semireverberant environment, to within 1 dB of the directly measured intensities.