Blind parameter identification is a research topic of high importance for both military and civilian communication systems. To our knowledge, there is no report in literature on blindly recognizing the modulation of noncooperative MIMO systems in association with space–time block code where channel state information and coding matrix are unavailable. In this paper, we first present a classifier based on maximum likelihood on the condition of virtual channel matrix; second, the modulations are classified into two classes according to the independence of the source signals: independent and groupwise independent constellations. In the next step, a multidimensional independent component analysis algorithm is proposed by utilizing the block-diagonal structure of the cumulant matrices to estimate the virtual channel matrix for these two cases, respectively. Last, the ambiguities are removed partly, and the classifier is proven to be insensitive to the remaining indeterminacy. Parallel computation technique is adopted to accelerate the computation of the logarithm likelihood function. Simulations show that our algorithm can work with high recognition probabilities in noncooperative space–time block code communication systems. Copyright © 2013 John Wiley & Sons, Ltd.