This paper deals with passive ambient seismic noise processing. A new wave separation method is proposed when several polarized waves are recorded by a pair of multicomponent sensors. Although it is well known that the information given by the auto-covariance matrix at one sensor is not sufficient for estimating the mixing matrix, it is shown in this paper that with two multicomponent sensors, the identification of the mixing matrix can be completed. In addition to the mathematical proof and synthetic example, an application to a real geophysical data set is also presented. A previous analysis of these data revealed the presence of surface waves. The application of the proposed algorithm leads to the separation of Rayleigh and Love waves.