Association among news articles is useful information for us to track situation related to events, persons, organizations and other concerned issues as well as to detect inconsistency among news. In this paper, we propose an association-based approach towards mining relations in Thai news articles by exploiting coincident terms. This approach first transforms news documents into term-document representation, applies term weighting techniques and generates association by means of statistics. In the work, either unigram or bigram is used for term representation, term frequency, boolean frequency and their modification with inverse document frequency are alternatively applied for term weighting, and confidence or conviction is in turn selected for association measure. Due to this combination, sixteen possible methods are investigated using approximately 811 Thai news of three categories, i.e., politics, economics, and crime. The ranked relations obtained by each method are compared with evaluation done by human. As the result, the method using bigram, term frequency, and conviction achieves the best performance with a rank-order mismatch of 0.84% on the top-50 mined relations. For the top-300 mined relations, the method with bigram, term frequency with inverse document frequency and conviction performs the best with 6.98% rank-order mismatch.