In the work the MODESLAB approach is applied to the sub-structural modeling of the anti-inflammatory activity of both natural and synthetic compounds, with the purpose of calculating the spectral moments of the adjacency matrix between edges of the molecular graph with suppressed hydrogens , weighted in the main diagonal with standard dipole moments of binding to 410 active and inactive compounds. The calculated descriptors were used in the design of a training series and another one of prediction. With the training series a discriminant function was developed for the anti-inflammatory activity by means of the Multivariate Linear Regression Discriminant analysis obtaining a good total classification of 91.59%. The model was validated through the use of the prediction series, obtaining a good classification of 90.2%.