The performance of the subspace-based noise power estimation algorithm for an OFDM receiver is investigated. It is shown that due to the correlation of the channel in the frequency domain, the dimension of the autocorrelation matrix does not need to be larger than the number of signal paths. Therefore, due to the reduction of the dimension of the signal space, the required complexity and computation for subspace decomposition is reduced. It is shown that the algorithm performs well in various speed and channel conditions.