Abstarct: Automatic Speech Recognition (ASR) consists in recognizing baxsed on a sample of speech from an unknown speaker.In order to recognize a speech signal. Any ASR systems need the feature extraction block. The Mel Frequency Cepstral Coefficient (MFCC) is one of the most common features which are used in ASR systems. The problem occurs when ASR systems is in the noisy environment because MFCC performance degrade drastically in noisy condition. In order to achieve a satisfactorily performance under noisy condition we need to improve the standard MFCC feature extraction method. In this thesis we introduce a noise robustness new set of MFCC vector estimated through some basic variation in the standard algorithm. We use the mean subtraction technique in both time domain and frequency domain, higher order autocorrelation coefficient are extracted and we apply eliminating the lower order of autocorrelation coefficient, using suitable filters to suppress convolution noise, using Gaussian shape filter bank in place of triangular shape filter bank and adding compensator block to enhance robustness of algorithm better. To evaluate the performance of proposed MFCC method and to classify the results we use MLP neural network with one input laxyer, two hidden laxyer and one output laxyer. We use forty isolated word spoken by twenty different speakers including male and female .frxame length is 44ms and sampling frequency rate is 22000. 70% of the entire data is used for train and 30% is used for testing. So far lots of approaches is used to improve MFCC algorithm some of them improve it through insisting on some basic block and some of the others improve it by adding complementary block to MFCC basic algorithm .In our proposed method we attend to most of last variations we use them in the best place and find the way to combine them in a best manner and we add some complementary block to standard algorithm too. Recognition experiments show significant improvement in recognition rate compare with standard MFCC algorithm and some popular algorithm in MFCC’s familes including: AMFCC، GMFCC، ROOT-MFCC،CMN-SMN-MFCC، RAS-MFCC