Between 2005 and 2013, four chemical tracer operations associated with short (2-3 months) to medium-term (5-6 months) circulation tests were conducted between the geothermal wells GPK-3 and GPK-2 of the EGS site at Soultz-sous-Forêts, in Alsace (France). The used tracers were 150 kg of fluorescein, 1.157 kg of SF6 gas, 200 kg of 1,3,5-naphthalene tri-sulfonate (1,3,5-nts) and 200 kg of 1,3,6-nts. During the 2005 circulation test, fluorescein injected into GPK-3 was also monitored in the fluid discharged from GPK-4, which was the other productive well. During the 2013 circulation test, a second tracer (200 kg of fluorescein) was also injected into GPK-4 used as another injector well. All these wells have been drilled at a depth of about 5000 m, where temperature is close to 200°C. This study compares and discusses the results and conclusions drawn from these tracer tests. If most of the data obtained in the tracer tests carried out between 2000 and 2005, and especially in the 2005 tracer test using fluorescein, are presented and commented by numerous authors in the literature, the next tests which were conducted after some stimulation and fluid circulation operations carried out after 2005 are less well known and interpreted. Among the main conclusions, we can notice that all the tracer tests gave evidence of fast and relatively direct hydraulic connections between GPK-3 and GPK-2 (short-scale fluid circulation loop with linear maximum velocities of about 4-8 m/h and predominant N-S fractures) at about 5000 m depth, comparable to those found from GPK-1 to GPK-2, in 1997, between 3500 and 3900 m depths, at a temperature close to 160°C. Except for SF6 gas, the existence of other larger and slower hydraulic connections between the wells GPK-3 and GPK-2 (large-scale and quasi-infinite fluid circulation loops with predominant NE-SW and NW-SE fractures) was also highlighted. Significant contributions of native geothermal brine (for which a flux was estimated at 1-1.2 m3/h) as well as relatively low rates of tracer recovery (< 30%) were confirmed in the fluid discharged from GPK-2. A SF6 gas trapping in the complexity of the porosity of the fracture system when the flow paths lengthen could explain the absence of contribution of the large-scale circulation loop in the shape of the restitution curve of this tracer as well as its very low recovery rate (< 2%) and the absence of dissolved SF6 in the geothermal fluid discharged from GPK-2. When they exist, the discrepancies found on the shapes of the tracer restitution curves and the associated hydrodynamic characteristics depend mainly on the nature of the tracer (gas or dissolved chemical compound), the fluid reinjection conditions (injection flow-rates and pressure) used for each circulation test and flow-rates of discharged fluid. The increasing values of the tracer mass recovery rates and tracer-swept fracture volumes estimated between the 2005 and 2010 tests (from 15.8 to 25.4% for the total tracer recovery rates and from 10,031 to 26,643 m3 for the total tracer-swept fracture volumes), using the same modelling methodology and taking into account the differences of discharged fluid volumes, were mainly interpreted as a possible improvement of the fluid circulation loops between GPK-3 and GPK-2, probably due to the stimulation and fluid circulation operations carried out after 2005, which induced an opening of the pre-existing connections. This improvement seems to be confirmed by the results obtained during the 2013 tracer test, but additional modelling tasks must be still performed. Contrary to the hydraulic connections between the wells GPK-3 and GPK-2, those existing between GPK-3 and GPK-4, or between GPK-4 and GPK-2, appeared to be very poor: - fluorescein was only detected in GPK-4 after 29 days of its injection into GPK-3 in 2005 and tracer recovery was lower than 2% at the end of the test; - fluorescein injected into GPK-4 in 2013 could have been detected at very low concentrations in the fluid discharged from GPK-2, after 60-73 days of its injection; - the linear maximum velocities for both tests were similar and close to 1 m/h.