Given their diverse dimensions and characteristics, learning games are difficult to evaluate. However, evaluation remains the only way to verify if the educational targets are being achieved and to detect any functional vulnerability inside the learning game. Based on a review of literature, in this paper we propose to structure the evaluation process in the design and experimental phases, using analytical and empirical evaluation methods in order to reduce the risk of a weak designed learning game. Indeed, we suggest that a well designed learning game should be useful from a pedagogical point of view, usable from a learner point of view and acceptable from the institution, the teacher/trainer and the learner points of views. For this purpose, we outlined four criteria classes that influence the evaluation process of such learning game. The first class criteria is related to the ludopedagogical environment which gathers learning and entertainment aspects of the learning game, the second class is related to the learner affective and cognitive reactions that are formed when using the learning game, the third class is related to the training context regarding to the curriculum and the learning games' operational conditions, finally, the fourth class is regarding to the learner profile which gathers information about his competencies background and preferences. After introducing the concepts of usability, usefulness and acceptability as well as the current analytical and empirical evaluation practices which exist in the field of game based learning, the analysis and measurement criteria are introduced before associating them with each of these three evaluation dimensions.