The concept of using more than the minimal set of the normal equations was introduced in spectral analysis in order to counteract the hypersensitvity of the model's parameters. A simultaneous vertical and horizontal extension of the autocorrelation matrix leads to an overdetermined, order-extended system. The application of the overdetermined approach to the analysis of short signal frames is examined. A comparative study of the autocorrelation lags used in the overdetermined matrices, based on the biased and unbiased autocorrelation estimators, as well as their estimation performance, is presented. The authors suggest the use of the biased measure in the case of low SNR (signal-to-noise ratio) and short data records, since experimental results illustrate a sensitivity of the unbiased measure to the noise ratio and the frame length. However, when the conditions become better (higher SNR or more samples available) the unbiased measure tends to yield more accurate estimates than those of the biased one.