Herein, we investigate the mismatch of the Kronecker structure based MIMO channel model. Finding the minimum mismatch is trivial and the result is always 0. The maximization problem, on the other hand, is not convex in general. Therefore we combine semidefinite programming with a grid search, to find a lower bound of the maximum mismatch. Numerical examples show that the bound is tight even for moderate number of points in the grid. The resulting worst case model error, is much larger than the model error previously observed in measurement campaigns, which confirms that the channel correlation matrix can, indeed, be well approximated by the Kronecker product of the correlation matrices at the transmitter and receiver for the measured NLOS indoor scenarios.