Abstr@~This [13,14]. The observational models are naturally homomorphic images of the operational models, but it is not so easy to find out the homomorphism induced by a family of experiments. Indeed, observational models cannot be defined by arbitrary homomorphisms ! A 'concrete' linear model for CCS is the operational model in which programs are rendered by sets of finite and infinite computations. An 'abstract' linear model for CCS is the observational model in which programs are identified if and only if no CCS experiment distinguishes between them. In spite of the effort on models for testing preorders [13,20], that very sensitive model has not yet been worked out, for it has been shown in [10] that some infinite experiments are strictly more sharp than any family of binary tests (finite or infinite). There lays the motivation for the present paper, where we produce the expected model for the full version of pure CCS.Let us proceed to a rapid review of some existing approaches on the side of abstract models, independently of the homomorphisms which induce them as quotients. First of all, one can distinguish between implicit quotient models and explicit models where explicit representations are dealt with [6] . Further separation may be observed between two classes of explicit models, namely the class of purely order theoretic models and the class of mixed space and order theoretic models. Classic ~-algebraic domains are most commonly used in models of the first class [7,13,19,20,24]. For the second class of models, the order relation is the reverse inclusion between closed subsets in some metric completion of languages, trees or tree-languages [11,18,27], and since the considered spaces have denumerable bases, the resulting complete partial (*) rR;gA -Camp,.,s ae r~ea.uL~eu _ F 35o4z RENNES CEDEK