Based on the spatial masking phenomenon and Ambisonics, a watermarking technique for audio signals is proposed. Ambisonics is a well known technique to encode and reproduce spatial information related to sound. The proposed method exploits this feature of Ambisonics. A watermark is represented as a slightly rotated version of the original sound scene. In another scenario in which Ambisonic signals are synthesized from mono or stereo signals, watermarks are embedded near the signal by adding a small copy of the host signal. This scenario presents the important advantage that reversible watermarking is possible if loudspeakers are arranged properly for playback. This advantage is attributed to the fact that first order Ambisonics represents audio signals as mutually orthogonal four-channel signals, thereby presenting a larger amount of data than in the original version. Formulation of the proposed method is explained. Listening tests demonstrate the relations between primary parameter values and imperceptibility of the watermark. Computer simulations conducted to assess the robustness of the proposed method against common signal processing attacks demonstrate that the method is robust against most of the tested attacks.