The main objective of this work is to' implement a simple isolated_word speech recognition system. The system compares a spoken word with the templates that have been created; if it finds a match, ft 'recognizes' the word, and the computer does whatever it has been set up to do when that word is recognized; i.e., the system will execute the DOS command which corresponds to the input speech signal. The vocabulary consists of 11 isolated words, which form some of the DOS commands, recorded for one male speaker. Linear Prediction Coding (LPC) using the autocorrelation method is the speech feature extraction approach used in this work. Vector Quantization (VQ) technique is used as a coding technique for the speech features vectors; each vector is assigned a symbol, which is an index number to a pre-determined set of reference vectors called a codebook. The feed-forward back-propagation neural network structure is proposed for constructing this system. The basic neural structure used in this research consists of three-layers. The proposed system is implemented on a Pentium Computer, 233 MHz speed, DOS Version 6.0 or higher, with sound blaster support, using Turbo Pascal language version 7.0.