Three quasi-dynamic pharmacophore models have been constructed for the complement inhibitor peptide compstatin, using first principles. Uniform sampling along 5-ns molecular dynamics trajectories provided dynamic conformers that are thought to represent the entire conformational space for nine training set molecules, compstatin, four active analogues, and four inactive analogues. The pharmacophore models were built using mixed physicochemical and structural properties of residues indispensable for structural stability and activity. Owing to the size and flexibility of compstatin, one-dimensional probability distributions of intrapharmacophore point distances, angles, and dihedral angles of different analogues spread over wide and overlapping ranges. More robust two-dimensional distance-angle probability distributions for two pharmacophore models discriminated individual analogues in terms of specific distance-angle pairs, but overall failed to identify the active and the inactive analogues as two distinct groups. Two-dimensional distance-dihedral angle probability distributions in a third pharmacophore model allowed discrimination of the groups of active and inactive analogues more effectively, with the highest-activity analogue having distinct behavior. The present study indicates that more stringent structural constraints should be used for a set of structurally similar but flexible peptides, as opposed to organic molecules, to convert dynamic conformers into pharmacophore models. Flexibility is a general aspect of the structure and function of peptides and should be taken into account in ligand-based pharmacophore design. However, the discrimination of activity using multidimensional probability surfaces depends on the peptide system, the selection of the training set, the molecular dynamics protocol, and the selection of the type and number of pharmacophore points.