This paper presents a multi-rank extension of the Capon beamformer. By expanding the rank of the beamformer it is possible to fully exploit situations in which signals lie in multi-dimensional subspaces, as opposed to the standard point source case. Such situations are commonly caused by array mismatches and scattered or distributed sources. The extension involves the design of a constraint matrix which can be interpreted in terms of signal power. Three possible choices for the constraint are proposed. These correspond to one non-adaptive choice, one choice that is dependent on the signal covariance structure only, and one choice that is both signal and data adaptive. Simulation examples are presented that show the promise of the idea of multi-rank Capon beamforming. Especially the signal- and data adaptive constraint appears very promising.