Technologically Enhanced Learning (TEL) potential has been demonstrated since the very beginning of computer science history [Skinner, 1965]. Two different points of view emerged: Skinner proposed to automate the transmissive tasks of the teacher; Piaget, on the contrary, proposed harness TEL to associate learners in the learning process according to the developmental learning theory (see [Kolb, 1984] and [Rezeau, 2001] for details). These two approaches were already deeply discussed in 1964 [Oléron, 1964]. In both theories, and since the very beginning, TEL environments were used to provide learning activities selected according to some learning strategies, and taking into account the productions of the learner and, in a more general way, his/her behavior. The goal was to “automatically” guide the learner throughout the learning process. In order to provide such guidance, these environments had to incorporate observation tools for monitoring learning activities. These tools were used to assess learners' progress and to guide the learning activities accordingly. Observations collected in TEL environments have to be modeled and represented so that they can be further reused by digital systems for various tasks such as assessment, diagnosis, adaptation, rewarding, panification, etc. It should be noted that the observation of a learning process goes way beyond TEL environments. Even in situations where no TEL are involved, teachers make observations, record them and reuse them for pedagogical purposes. Even if the observation of a learning process has always played a central part in pedagogical practices, research on that topic has not always been very active. For some years now, observation and dynamical (and sometimes real-time) exploitation of these observation is a research question of a growing importance. With the multiplication of MOOCs (Massive Online Open Courses), this research question becomes even more important and raises new challenges such as the ability to observe thousands of learners involved in the same learning activity and to build relevant knowledge and services based on these observations. Among these elements, we can list: indicators, dashboards, learner profiles, etc. Learner profiles are often built for teachers and tutors, but some systems make an effort to make their semantics clear for learners too (see for example The Observer 1 ). There is a very significant literature on the subject of making easier the exploitation of this knowledge in the classroom, specifically in connection with learners profiles engineering [Brusilovsky & Millan, 2007], [Ginon et al, 2011]. In this paper, we address “observation” as a research question. More precisely, we study the concepts underlying instrumentation, collection and representation of observations in web based TEL environments. We develop models and tools according to this conceptualization. Obviously, observation and interpretation are strongly related. In the literature, it is generally admitted that “interpretation knowledge” is an abstract knowledge. Thus, the question raised is “how to connect this abstract level knowledge to low level observations?”. Establishing such a connection often lead to the implementation of top-down ad hoc observation processes, deeply integrated in TEL environments. Top-down approaches for implementing observation mechanisms are rather complex to implement (see recommendations of the Alberta University for building profiles 2 ), and some researchers, such as [Choquet & Iksal, 2007], propose to integrate specification of what has to be observed in the authoring process in order to elaborate learning indicator.