A new noise variance and SNR estimation algorithm for a 2 × 2 MIMO wireless OFDM system as defined in the IST-STINGRAY project is presented. The SNR information is used to adapt parameters or reconfigure parts of the transmitter. The noise variance estimation algorithm uses only 2 OFDM training symbols from each transmitting antenna and the FFT output signals at the receiver. It does not require knowledge of the channel coefficients. Then, using the channel coefficient estimates given by a channel estimator and the estimate of the noise variance, the SNR is computed. The algorithm's performance is measured through Monte-Carlo simulations on a variety of channel models and compared to those of an MMSE algorithm using perfect channel estimates. The normalized MSE of the obtained noise variance estimate shows good results as long as the delay spread of the channel is small enough compared to the OFDM symbol period.