Gas cooling in nuclear power plants (NPPs) has a long history, the corresponding reactor types developed in France, the UK and the US having been thermal neutron-spectrum systems using graphite as the moderator. The majority of NPPs worldwide, however, are currently light water reactors, using ordinary water as both coolant and moderator. These NPPs – of the so-called second generation – will soon need replacement, and a third generation is now being made available, offering increased safety while still based on light water technology. For the longer-term future, viz. beyond the year 2030, R&D is currently ongoing on Generation IV NPPs, aimed at achieving closure of the nuclear fuel cycle, and hence both drastically improved utilization of fuel resources and minimization of long-lived radioactive wastes. Since the very beginning of the international cooperation on Generation IV, viz. the year 2000, the main research interest in Europe as regards the advanced fast-spectrum systems needed for achieving complete fuel cycle closure, has been for the Sodium-cooled Fast Reactor (SFR). However, the Gas-cooled Fast Reactor (GFR) is currently considered as the main back-up solution. Like the SFR, the GFR is an efficient breeder, also able to work as iso-breeder using simply natural uranium as feed and producing waste which is predominantly in the form of fission products. The main drawback of the GFR is the difficulty to evacuate decay heat following a loss-of-coolant accident (LOCA) due to the low thermal inertia of the core, as well as to the low coolant density. The present doctoral research focuses on the improvement of decay heat removal (DHR) for the Generation-IV GFR. The reference GFR system design considered in the thesis is the 2006 CEA concept, with a power of 2400 MWth. The CEA 2006 DHR strategy foresees, in all accidental cases (independent of the system pressure), that the reactor is shut down. For high-pressure events, dedicated DHR loops with blowers and heat exchangers are designed to operate when the power conversion system cannot be used to provide acceptable core temperatures under natural convection conditions. For depressurized events, the strategy relies on a dedicated small containment (called the guard containment) providing an intermediate back-up pressure. The DHR blowers, designed to work under these pressure conditions, need to be powered either by the power grid or by batteries for at least 24 hours. The specific contributions of the present research – aimed at achieving enhanced passivity of the DHR system for the GFR – are design and analysis related to (1) the injection of heavy gas into the primary circuit after a LOCA, to enable natural convection cooling at an intermediate-pressure level, and (2) an autonomous Brayton loop to evacuate decay heat at low primary pressure in case of a loss of the guard-containment pressure. Both these developments reduce the dependence on blower power availability considerably. First, the thermal-hydraulic codes used in the study – TRACE and CATHARE – are validated for gas cooling. The validation includes benchmark comparisons between the codes, serving to identify the sensitivity of the results to the different modeling assumptions. The parameters found to be the most sensitive in this analysis, such as heat transfer and friction models, are then validated via a detailed re-analysis of earlier PSI (EIR, at the time) gas-loop experiments conducted in the 1970s. Conclusions and recommendations on the models to be used for transient analysis are derived. In general, it has been shown that the agreement, between experiments and the correlations for heat transfer and friction used in TRACE and CATHARE, is quite satisfactory. The thus validated codes are then used in the two detailed, DHR improvement studies carried out. The first improvement of the reference DHR strategy is the heavy gas injection. Assuming a DHR blower failure after a LOCA, the helium pressure in the guard containment is not high enough to evacuate the decay heat by natural convection. To improve the natural convection, the effects of injecting different heavy gases (N2, CO2, Ar and a N2/He mixture) into the primary circuit were analyzed, in order to address the possibility of dealing with DHR-blower failure while accepting an intermediate back-up pressure in the guard containment. Furthermore, different injection locations and injection mass flows were considered, and the sensitivity to the number of available DHR loops and LOCA break-sizes was also addressed. It has been found that injecting the heavy gas in the vicinity of the core could lead to overcooling problems. For an injection point sufficiently far from the core, however, both CO2 and N2 are found to be able to cool the core satisfactorily in natural convection. N2 is proposed as the reference, due to possible chemical problems with CO2. The second proposition for DHR improvement is related to the possibility of a simultaneous guard-containment failure, i.e. a loss-of-back-up-pressure (LOBP) combined with a blower failure after a LOCA. In this case the natural convection, even with heavy gas injection, is no longer strong enough to evacuate the decay heat. To address this issue, the possibility of decay heat removal via use of a dedicated autonomous Brayton cycle – as a standalone DHR loop – has been investigated. First, an analytical Brayton cycle model has been set up, so as to identify convenient machine design points and to study the machine's off-design behavior. Two machine designs have then been drawn up: one for helium in order to provide a reference for understanding the Brayton loop behavior in a generic sense, and the other for nitrogen which is the envisaged gas to be injected after a LOCA. Both, the design of the proposed devices and their validation are discussed. Finally, a detailed transient analysis, involving usage of both heavy-gas injection and the Brayton device (i.e. of the complete, proposed DHR system), is presented. This serves to illustrate the effectiveness of the new strategy for the highly hypothetical worst-case scenario of sequential failures following a LOCA.