Most of the existing usability evaluation and testing methods require a fully functional prototype. As a consequence, tests are conducted after the development and most often, after the deployment of the whole software. Furthermore, tests require a costly usability laboratory and highly trained usability testers, usually developers lack training in conducting such tests. Cost-benefit studies show that these problems and similar ones result in significant costs. Predictive usability models have been introduced as potential solutions to address these crucial drawbacks of the existing usability evaluation methods. Predictive usability models and measures can supplement the existing evaluation methods while reducing costs and enhancing efficiency, accuracy and objectiveness of the tests.  In this thesis, we demonstrated via empirical investigations, that usability measures and user-oriented tests conducted by users can provide similar scores regarding the overall usability as well as two core usability parameters: Learnability and Efficiency. Moreover this thesis demonstrated that the results of empirical investigations can be used to build measure-based models for usability prediction. As an outcome, this thesis introduced a comprehensive methodology to develop and validate measure-based models for usability prediction from early user interface design artifacts including storyboards and prototypes. This methodology includes a systematic process that involves discovery of correlations between usability measures and the results of usability tests performed by users.