An audio source localization apparatus receives signals from a microphrine array (101), and a reference processor (105) generates at least three reference beams with different directional properties. An estimation processor (107) which generates simultaneous direction estimates for two sound sources, comprises a circuit (401) combining signals of the at least three reference beams with a beam shape parameter reflecting a shape of an audio beamform and a beam direction parameter reflecting a direction of an audio beamform for the combined signal. A cost processor (403) generates a cost measure indicative of an energy of the combined signal and a minimization processor (405) estimates values of the beam shape parameter and the beam direction parameter which correspond to a local minimum of the cost measure. A direction processor (407) then determines simultaneous direction estimates for two sound sources from the determined parameter values. Improved direction estimation for two simultaneous sound sources may be achieved.