Speech recognition can simply be defined as the process of transforming acoustic speech signal into discrete representation, which defines the information carrying features. The recognition process could be performed on Words, Phones, Allophone, Diaphone (explicit transition modeling), Syllables (multi/phone units), and Triphones (context-dependent phones). This work is concerned with developing speech recognition system for an English sentence (consist of words pronounced with one second time interval). At first the sentence is split into words (depending on energy and zero crossing calculations), then Linear Predictive Code (LPC) is applied on each word to extract its features. The resulted features are quantized using Linde-Buzo-Gray (LEG) to perform feature coding (useful in reduction of recognition time). Finally, feedforward neural network (Back-Propagation) is used as a classifier. The proposed system is tested using five different sentences (17 words) pronounced by single person. The evaluation of the distance measure between the test and reference patterns is implemented by Euclidian distance and the result for this method gives an acceptable accuracy.