Terahertz spectroscopy provides a new way for material identification. Investigation on establishment and usage methods of terahertz spectral database is needed so as to distinguish unknown substances. In order to establish the database, terahertz spectra of twenty organic materials are measured by own built terahertz time-domain spectroscopy system and the interference information of baseline and noise are removed by wavelet transform. The usage of database is divided into two steps: 1) determine whether the unknown substance is in the database through the radial basis function neural network; 2) identify the material if it is in the database by multi-class support vector machine. The fault tolerance of the algorithm is improved combined with error-correcting output coding to handle the multi-class problem with recognition rate of 96.7%. The network also has a good prediction of materials outside database with recognition rate of 93.2%. The establishment and usage methods of the database suppress the system noise and can be applied to practical situations.