Synthetic aperture magnetometry (SAM) is a nonlinear beamformer technique for producing 3D images of cortical activity from magnetoencephalography data. We have previously shown how SAM images can be spatially normalised and averaged to form a group image. In this paper we show how nonparametric permutation methods can be used to make robust statistical inference about group SAM data. Data from a biological motion direction discrimination experiment were analysed using both a nonparametric analysis toolbox (SnPM) and a conventional parametric approach utilising Gaussian field theory. In data from a group of six subjects, we were able to show robust group activation at the P < 0.05 (corrected) level using the nonparametric methods, while no significant clusters were found using the conventional parametric approach. Activation was found using SnPM in several regions of right occipital-temporal cortex, including the superior temporal sulcus, V5/MT, the fusiform gyrus, and the lateral occipital complex.