FTART(field theory—based ART)algorithm combines the theory of ART (adaptive resonance theory)，ARTMAP and field theory．It corrects the generated classi-fication regarding the supposed distribution possibility of examples in the instance space．FTART employes a different conflict resolve process and dynamicly expanding the classifi-cation area．By further studies，the paper improves the algorithm'S ability of learning continuous function mappings．Some benchmark test results and the analysis are also given．They proved that FTART has good performance on pattern recognition and continuous function mapping．