In search of new features having complementary information, this paper investigated the use of entropy of the spectrum as additional feature. It has been shown that entropy of the full-band spectrum can be used as an estimate of voicing/unvoicing. Going one step further we suggested dividing the spectrum into equal sub-bands and obtaining entropy from each sub-band and using that as an additional feature for ASR. Good improvement in performance is obtained when multi-band entropy feature is appended to the usual PLP cepstral features, specially in case of noise. The new feature though doesn’t compete with the cepstral features in absolute sense, it seems to be quite robust to noise. The reason for robustness can be attributed to the fact that multi-band entropy feature tries to capture the location of the formant and formant are less affected by noise.