A multivariate 'forensiometric' technique (PLS-DA) was used to create a model to predict the relationship between homicide victim and perpetrator on a five-level scale: strangers, acquaintances, drinking companions, relatives and spouses.The model is based on findings on the victim and at the venue, and uses the following ten variables as 'predictors', listed in falling order of covariation with instances of close relationship between victim and perpetrator: Victim found in home, female victim, a single sharp injury, injuries to the upper extremity, superficial sharp injuries to the chest ('scratches'), ten or more sharp injuries, presence of defence injuries, total number of sharp injuries, male victim, and victim found outdoors.The model was based on 87 sharp-force homicides (the model-set) and was validated on another set of 43 sharp-force homicides (the test-set). In this test-set validation, 17 of 39 cases (44%) were correctly predicted according to the results of earlier combined forensic and police investigations, and a prediction one step away from the correct level was given in another 17 cases. These results were significantly better than would have been obtained by chance.