Diabetes mellitus is a globally growing disease with more than 180 million cases worldwide and no cure exists. In order to minimize the many medical complications, tight glucose monitoring has been shown to be the best alternative. Many diabetics require insulin injections to regulate their glucose level and, as of today, the insulin quantity to be delivered can only be determined on the basis of finger stick measurements. Such measurements are painful and insufficient, and therefore continuous glucose monitoring systems are necessary. A few devices are recently available for 2 - 7 days of use, but since these devices adopt a semi-invasive principle, they are not suited for long term (e.g. 1 year) monitoring. In practice, it is necessary for a long term sensor to be implanted and such a device does not exist yet. A novel method has been recently developed in our laboratory to continuously monitor the glucose concentration in the subcutaneous interstitial fluid. The method uses an implanted rotating microviscometer to measure the viscosity variations of a glucose-sensitive fluid due to the glucose concentration changes. The present work investigates solutions for a biocompatible interface between the living tissues and the sensitive fluid enclosed in the device. Because the sensitive fluid contains two necessary molecules, dextran and concanavalin A, the interface has to selectively retain these two solutes whilst being permeable to glucose. On the basis of the requirements of the sensor, a wide literature search and theoretical considerations, two selective interfaces have been selected among the most actual technologies: hybrid membranes composed of a nanoporous alumina membrane coated with poly(poly(ethylene glycol) methacrylate) (PPEGMA) brushes and nanoporous polyethylene films, which have different advantages from each other. A theoretical model for the transient diffusion of a solute through membranes permitted to quantify the membrane performance. Both selective interfaces have been characterized systematically and their selectivity to a test protein has been demonstrated. Furthermore, the link between the diffusion rate of glucose and the protein retention threshold has been highlighted, which enabled to determine the membrane performance without complex analysis. The optimized hybrid porous alumina-PPEGMA selective interface revealed sufficient retention properties of dextran and ConA in diffusion cells over 48 hours. In a second step, the hybrid selective interface has been integrated to a latest implantable sensor demonstrator and the interface performance has been investigated under an environment mimicking the in vivo conditions. Since the integrated demonstrator showed a good ability to respond to glucose variations in vitro, the system has been transferred for in vivo experimentation. An in vivo test has been carried out over 5 days in a rat model, and the implantation as well as the transcutaneous data acquisition have been successfully demonstrated. Finally, the sensor capability to respond to glucose variation in a rat model has been shown and solutions for further development of the device are proposed.