A series of tirucallane triterpenoids isolated from the stem barks of Aphanamixis grandifolia were assessed for their anti-inflammatory activity, exhibiting from weak to strong anti-inflammatory activity, by testing their inhibitory effects on nitric oxide (NO) production and tumor necrosis factor-α (TNF-α) level in lipopolysaccharide (LPS) induced RAW264.7 murine macrophages. To explore the relationship between the structures and anti-inflammatory activity, a vast pool of molecular descriptors for each isolate were calculated. Genetic algorithms (GA) or simulated annealing (SA) based partial least squares (GA-PLS and SA-PLS) algorithms identified some important descriptor combinations, which were correlated with both sets of the anti-inflammatory data by partial least squares 2 (PLS2) method. S-Plot was used to visualize the descriptor influence in the PLS2 model, disclosed five most important molecular descriptors, nOHt, RDF150m, lip_violation, Mor32m and JhetZ.