A new type of 3D-QSAR descriptors is introduced. For each molecule under consideration an internal coordinate system is defined relative to molecular points, such as positions of atoms in the molecule or centers of mass or certain substructures. From the origin of this system distances to the solvent accessible surface are calculated at defined spherical coordinate angles, theta and phi. The distances represent steric features, while the molecular electrostatic potentials at the intersection points with the surface represent the electrostatic contributions. The approach is called IDA (internal distances analysis). Matrices obtained by varying the spherical coordinate angles by fixed increments are correlated with the biological activity by partial least squares (PLS). The descriptors, tested with the benchmark steroids and an also well characterized benzodiazepine data set, turn out to be highly predictive. Additionally, they share the advantage of grid-based methods that the obtained models can be visualized, and thus be directly used in a rational drug design approach.