First-order electron density similarity measures for large molecules are straightforward and can be efficiently computed if the atomic shell approximation (ASA) is used. Within this approximation the molecular electron distributions are represented by simple superpositions of spherical atomic contributions. A new algorithm to optimally select shells fitting known electron distributions and an empirical scheme to construct molecular densities by summing atomic fragments are presented. The accuracy of both ASA procedures is analyzed comparing approximated and ab initio QMSM.