Can we - software developers, usability experts, user interface designers - predict usability from the early user interface (UI) design artifacts and models? Can we define predictive measures to evaluate usability without a concrete UI? These questions seemed natural for us since UI modeling (task, user, concepts, etc.) is being largely explored in recent years for the automatic generation of final UI. To answer those questions we propose a model-based predictive usability evaluation approach that uses a set of usability measures. These measures are the essence of a framework we are developing for usability prediction. Initial empirical studies were performed to support this approach. This paper presents the fundamental basis on top of which we have developed this approach.