This paper proposes a method for detecting entity mentions in the given text. The method consists of two sequential labelling steps. The first sequential labelling is responsible for identifying a chunk of words (morphemes) mentioning an entity. The second sequential labelling verifies whether or not each chunk (or un-chunked word) is really an entity. The paper also investigates the relation between granularity of categories of named entities and mention detection performance. Our experimental results have shown that the second sequential labeling step for checking slightly improved recall and the eleven categories is the best granularity.