The 3D reconstruction of facial features from skeletal remains is a key component to the identification of missing persons and victims of violent crime. A comprehensive Computed Tomography (CT) head-scan database is currently being collected which will enable a new approach to forensic facial reconstruction. Using this unique resource, we show how a face space can be tailored to a specific unknown, or questioned skull. A set of database derived estimates of the questioned face is constructed by first computing non-rigid transformations between the known head-scan skulls and the questioned skull followed by application of these transformations to the known head-scan faces. This effectively factors out influences due to skeletal variation. A tailored face space is formed by applying Principal Component Analysis (PCA) to this ensemble of estimates of the questioned face. Thus, the face space is a direct approximation of correlated soft tissue variance indicative of the population. Ours is the first mathematical representation of the face continuum associated with a given skull. Embedded in this space resides the elements needed for recognition.