Abstarct: Detecting Keywords or Keyword Spotting, generally means finding a keyword in a written or spoken document. In this research, a new approach for detecting or recognizing keywords in Persian language is proposed, to spot keywords in continues and discrete utterances. Dynamic Time Warping approach is used in the both parts which is different from Hidden Markov Model baxsed approaches that is widely used today. The proposed algorithm for keyword spotting in continues utterance is baxsed on a modified version on DTW, named Segmental DTW. DTW is a classic distance measure for calculating the similarity between two sequences which are different in length. In the processing stage, the utterance is divided into short frxames. Each frxame is represented in forms of quantized feature vectors. Both keywords and utterance are converted into a sequence of indices of codebooks. After that, the sequence of codebook indices of the utterance is divided into some segments. Later on, the system calculates the DTW distance of each of segment with query keyword(s). finally the segment with the minimum of distortion score, is most likely to be a keyword. We also have proposed an approach for keyword spotting in isolated words. In this part, we first prepare a dataset from desired keywords. We compose a generic model for any of desired keywords, using alignment path in Dynamic Time Warping algorithm. In the test stage, we calculated DTW the distance between generic models of keywords and the incoming words. The model with the minimum distortion score is candidate to be a keyword.