Speech is very important for communication. With the help of speech one can express their thoughts, interact with people easily. Feature Extraction is the very first step in speech recognition. This paper presents the effect of combination of different features on speech recognition. Different features of speech such as spectral, temporal, perceptual and energy. Also accuracy for each feature is calculated and it seems that Mel Frequency Cepstral Coefficients (MFCC) gives better accuracy among all the features. And the accuracy of MFCC is further increased by combination of MFCC with bark.