This paper describes the Subband-Autocorrelation (SBCOR) analysis technique and investigates how to apply it to speech recognition. The SBCOR analysis is a new signal analysis technique based on filter bank and autocorrelation analysis. The SBCOR analysis system is evaluated for five types of filter bank and three autocorrelation detectors using a speaker-dependent DTW word recognition system. The experimental results show that the SBCOR spectrum performs equally as well as the smoothed group delay spectrum under clean conditions, and much better than it under noisy conditions. Finally, it is shown that the most suitable filter bank is a fixed Q filter bank whose center frequencies are equally spaced on the Bark scale, and the most suitable autocorrelation analysis is a conventional autocorrelation detection without controlling weak signals. An analysis example of speech is also shown under these conditions.