Recently, the prediction of promoters has attracted many researchers' attention. Unfortunately, most previous prediction algorithms did not provide high enough sensitivity and specificity. The goal of this thesis is to develop an efficient prediction algorithm that can increase the detection power (power = 1 - false negative). We do not try to find more distinct features in promoters one by one, such as transcriptional elements. Our main idea is to use the computer power to calculate all possible patterns which are the possible features of promoters. Accordingly, we shall define some scoring methods for training a given set of sequences, which involve promoter sequences and non-promoter sequences. Then, we can obtain a threshold value for determining whether a testing sequence is a promoter or not. By the experimental results, our prediction has higher correct rate than other previous methods.