A system for the extraction of uncorrelated reverberation from two-channel (stereo) audio signals is proposed and evaluated. Applications for the new system vary from surround-sound multichannel loudspeaker upmixers for home-theater or automotive audio systems, to headphone-based auralization for enhancing the spatial sound quality of a listening experience. The new system uses the normalized-least-mean-square (NLMS) algorithm to equalize the two input signals with respect to both spectral magnitude and phase before differencing to remove correlated components. A theoretical model of the system based on a stochastic room impulse response model was validated by empirical measurements made in a reverberant hall with a microphone pair, and from a formal subjective evaluation the system is shown to be an effective approach to extracting reverberation from audio recordings.