In order to investigate the clutter nulling performance of the sample matrix inversion (SMI) algorithm in amplitude heterogeneous environments, we derive an analytical expression for the average signal-to-interference plus noise ratio (SINR) loss based on random matrix theory. The results show that comparing with the case where the training data utilizes the homogeneous secondary samples (which are independent and identically distributed (IID) to the snapshot in the cell under test (CUT)), formulating the adaptive weight with the secondary samples which have stronger clutter power than that in CUT will increase the output SINR. Nevertheless, the achievable performance improvement might be quite limited. Conversely, selecting the secondary samples with weaker clutter power will degrade the performance.