Relevance analysis is a regular and important task in many technical fields. We can get the relevance score by measuring (or quantifying) the result of relevance analysis. In this paper, we have reviewed two main tools for relevance measure, which are the covariance and the mutual information, and we have discussed that there may be some problems in relevance measure if we use the above two methods, then we give the definition on Partial Condition Entropy(PCE) based on the information theory and presented a new method for relevance measure by using the PCE. There are mainly three advantages for relevance measureby using our method: (1) The relevance degree can be compared more easy than other methods because the score of relevance calculation is equal to a numeral between 0 and 1;(2) By using the method, we can not only know whether there is relevance between the considered events but also get a special score that represents the relevance degree of these events; (3) When we calculate the PCE, we needn't know all the conditional probability density, so our method is more flexible than the calculation of mutual information. To demonstrate the usefulness of our method for relevance measure, we apply it to the sentence relevance analysis in Natural Language Processing(NLP). We find that our result of relevance measure is a more truly reflection on the relationship between the sentences.
