The aim of this study was to compare the efficiency of two well known approaches for the discovery of the bioactive principle/s in medicinal plants, namely the activity-guided isolation versus the computer-aided drug discovery by means of virtual screening (VS) techniques. Morus root bark of Morus sp. L. (Moraceae) was selected as application example for the discovery of compounds with anti-inflammatory activity. The two cyclooxygenase isoenzymes COX-1 and COX-2 were chosen as targets and the corresponding pharmacophore models were generated by our research. The activity-guided fractionation of the methanol extract of the root bark resulted in the isolation of nine compounds. Their structures were elucidated by mass spectrometry, 1- and 2-dimensional NMR experiments and identified as moracins B, M, the regioisomers O/P as a mixture, and sanggenons B, C, D, E and O. The COX-1 and COX-2 inhibiting activities of these compounds were established in an enzyme assay and compared with the predicted hits obtained from the VS. Sanggenons C, E, and O, that were tested the first time for an inhibitory effect on COX-1 and -2, showed IC50 values of 10-14 microM, and 40-50 microM, respectively. The results show that the COX activities obtained for the sanggenons are correctly predicted by the in silico filtering experiment. In the case of the isolated moracins, however, it failed because the COX inhibiting activities of moracins M and P/O were not retrieved by the VS. Structure-activity relationships of the isolated compounds are discussed as well as potential pitfalls and advantages of the applied strategies.