Abstarct: Conversation is a type of two-way communication between two humans. From the perspective of psychologists, a correct conversation has features that one of them is coherence. Coherence in conversation means keeping path, lack of gaps and sudden changes in the subject of the conversation. Changing the subject in a coherent conversation happens as fluent and relevant to past subjects. A conversation can be connected between humans and system, such as dialogue and interactive question answering systems. In order to increase the efficiency and popularity of the dialogue systems, the design of these systems should be such that the correct conversational criteria between the two humans are met. In fact, the dialogue systems must be respond, like an expert human. The correct conversation between two humans has the characteristics that their observance in dialogue systems, can increase coherence. Features, are identifiable by comparing coherent and non-coherent conversations. By recognizing the characteristic in coherent and non-coherent conversations, can be measure the coherence of a conversation. So far, in built-conversational systems, coherence have been observed on them. The designers of these systems are used human opinions for to check the coherence of conversations. Use of human opinion is useful, but human cannot give the score to the systems, with the same opinion. So far, the automated system has not been designed, to be able give scores to the conversations with a same opinion. In this research, by investigating the coherent and non-coherent produced conversations by human, the features, have been suggested to assess the level of coherence of a conversation. The features are divided into two groups of repetition and collocation of keywords. Each category of introduced features has been tested on the conversation series. The F measure derived from the conversation categorization with related features to key words repetition with support vector machine is 62.8%. The combination of the repetition and the collocation of the key words, has led to the F measure to 64%.