Abstract: This paper establishes a Mandarin emotional speech database that combines discrete emotion tags with dimensional emotion space. The database is recorded for 16 Chinese native speakers in performing Chinese emotional speech. The speech samples are acquired from seven discrete emotion tags, such as neutrality, pleasure, happyness, frustration, anger, sorrow, and sadness. Each speaker receives 336 utterances. Then, each of the speech samples is annotated by three annotators in dimensional space. Finally, according to the obtained data, the dis-tributions of these seven emotions in the emotion space are studied, and the performances in consistency, concentration and difference of these emotions are analyzed. Besides, we calculate the emotion recognition rates of these seven emotional speech. The analyses show that the consistencies of the three annotators for the database are more than 80%, and these emotions can be distinguished, in addition, the recognition rates of these seven emotions are all higher than baseline level. Therefore, the database has a good emotional quality, and can provide important research basis for the transformation of discrete emotion tags to dimensional emotion space.