The adaptive optics system of the 6.5m MMT, based on a deformable secondary mirror has been of the sky now for 6 runs of roughly 2 weeks each. Altogether its performance has been quite satisfactory with a crop of science results. However, even if the mirror has shown it promises, it has proven difficult to improve the system in terms of wavefront quality much beyond what it achieved during 1st light. In particular, it has not been possible to improve image quality by using a larger number of modes than the 52 modes used originally. One reason for this behavior could be that the interaction matrix used to build the AO controller has been measured in lab conditions using a different optical set-up than the one on the telescope. In a Cassegrain adaptive secondary AO system measuring the interaction matrix in the way all other AO systems do is impossible when the secondary is mounted on the telescope. In this paper, we present a way of measuring the interaction matrix on the telescope using real natural guide stars as sources. We present simulations results that show that this method can be used to measure the interaction matrix to a high signal-to-noise ratio.