A method for accurately and efficiently determining the location of points or entities in a machine vision image. A set of sub-models are constructed from a global training model representing objects viewed by the machine vision system. The set of sub-models are used to fix the actual location of user selected points or entities. Configuration of the sub-models is automated to ensure that the sub-models contain sufficiently stable features within optimally sized regions without requiring substantial user expertise. Located entities are compared to evaluate compliance with pre-selected tolerances.