This paper describes an extension of snbband-autocorrelation (SBCOR) analysis that has been already proposed and shown to be robust against noise. The extended SBCOR analysis is defined by a weighted sum of the autocorrelation coefficients at the integral multiples of 1/CF (center frequency of band pass filter), in order to capture more information about the periodicity included in the speech signal. The experimental results performed by a DTW word recognition indicate that the extended SBCOR analysis is more robust against noise than the conventional SBCOR. The performance of the extended SBCOR is about 8% higher than that of the conventional SBCOR(when Q=1.5), and about 20% higher than that of the smoothed group delay spectrum under SNR OdB. The characteristics of the SB-COR and its extension are also described.