To evaluate the performance of crosstalk cancellation systems the channel separation is usually used as parameter.  However, no systematic evaluation of the minimum audible channel separation has been found in the literature known by the authors.  This paper describes a set of subjective experiments carried out to evaluate the minimum amount of  channel separation needed such the binaural signals with crosstalk are perceived to be equal to the binaural signals reproduced without crosstalk.  A three alternative-forced-choice discrimination experiment, with a simple adaptive algorithm with weighed up-down method was used.  The minimum audible channel separation was evaluated for the listeners placed at symmetric and asymmetric positions with respect to the loudspeakers.  Eight different stimuli placed a two different locations were evaluated.  Span angles of 12 and 60 degrees were  also simulated.  Results indicate that  in order to avoid lateralization the  channel separation should be below -15dB for most of the stimuli and around -20dB for broad-band noise.