Brain source localization accuracy is known to be dependent on the EEG sensor placement over the head surface. In Brain-Computer Interfaces (BCI), according to the paradigm used, Motor Imagery (MI) and Steady-State Visual Evoked Potential (SSVEP) in particular, electrodes are not well distributed over the head, and their number is not standardized as in classical clinical applications. We propose in this paper a method for quantifying the expected quality of source localization with respect of the sensor placement, known as EEG montage. Our method, based on a subspace correlation metric, can be used to assess which brain sources can be distinguished (as they generate sufficiently different potentials on the electrodes), and also to identify regions/volumes in which precise source localization is impossible (i.e. all sources inside those regions could generate similar electrode potentials). In particular, for a MI dedicated montage, we show that source localization is less precise than for standard montages, although the local density of electrodes over the areas of interest is higher.