Based on the background constructing, peak recognizing and profile decomposing algorithms recently proposed by the authors, an expert system has been developed for fully computerized qualitative interpretation of wavelength-dispersive X-ray fluorescence spectra. The main contributions made in the development of this system are: (1) a new data structure, called peak-line-element net (PLEN), is proposed to construct a dynamic data base for expressing the complex relations between peaks, characteristic lines (CLs) and possible elements; (2) a procedure which arranges CLs according to their relative importance and eliminates obvious wrong matches is suggested in order to simplify the inference process; (3) a certainty factor (CF) which possesses the property of probability is defined by considering the overall matching degree of CLs. The expert system developed in this study was tested with seven samples of different types and origin, and on average its interpretation agreed with human experts for more than 90% of the elements involved.