This paper describe about the development of word recognition system by using a voice matching approach, which will match between reference voice spectrum and unknown voice spectrum. This designed system is limited to recognize a voice, which only consists of one word. There are five words, which will be recognized in this system. They are "satu", "dua", "tiga", "empat", and "lima". Input device of the system is a microphone. The output of microphone will be amplified by using a pre-amplifier circuit. Then, the signal is converted to digital signal by using an analog to digital converter (ADC), so that, the microprocessor TMS320C5x can read the data from the signal. This system uses linear predictive coding (LPC) to extract the pattern of the voice. Word recognition process is done by using nearest neighbor classifier method. Basic principle of this method is to find the shortest distance between unknown voice pattern and reference voice patterns. An unknown voice pattern is recognized as the reference voice pattern, which has the shortest distance with that unknown pattern. Some experiments have been done in various people voices. They are people whose voices are in reference voice database, people whose voices are not in database. Experimental results show that the system can recognize the word up to 92 %.