It is well known that the performances of any receiver can be improved by means of equalization and diversity combining. However, the problem of the blind adaptation is entirely open in the case of decision feedback equalization. This paper attempts to introduce a new scheme of blind spatio-temporal equalizer whose main feature is its ability to self-adapt both its structure and its criteria to the encountered situations. Simulations results are given for a 4-QAM signal and compared to those of the classical constant modulus algorithm (CMA) spatio-temporal equalizer. The results are really convincing : it clearly appears that the new equalizer exhibits very good performances in terms of mean square erro, bit error rate as well as convergence speed. From all these points of view, it largely outperforms the classical CMA linear equalizer. As a result, this new blind spatio-temporal equalizer turns out to be very attractive for many applications.