Structure-activity relationships study was performed for a few series of cyclooxygenase-2 (COX-2) inhibitors by using the Electronic-Topological Method combined with Neural Networks (ETM-NN). Specific molecular fragments were found for active compounds ('activity features') from both series by the ETM application. After this, a system of prognosis was developed as the result of training Kohonen's self-organizing maps (SOM) by the fragments. From the detailed analysis of all compounds under study, requirements necessary for a compound to be COX-2 inhibitor were formulated. The analysis showed that any requirements violation for a molecule resulted in a considerable decrease or even complete loss of its activity. The found activity features identified correctly different marketed drugs and new compounds that had passed pre-clinical and clinical trials; this fact confirms the workability of the system developed for the COX-2 inhibitory activity prediction.