We present the results of a detailed time resolved luminescence study carried out on a very high quality InGaAs quantum well sample where the contributions at the energy of the exciton and at the band edge can be clearly separated. We perform this experiment with a spectral resolution and a sensitivity of the set-up allowing to keep the observation of these two separate contributions over a broad range of times and densities. This allows us to directly evidence the exciton formation time, which depends on the density as expected from theory. We also evidence the dominant contribution of a minority of excitons to the luminescence signal, and the absence of thermodynamical equilibrium at low densities.PACS numbers: 71.35.Cc,71.35.Ee,73.21.Fg,78.47.+p,78.67.De Excitons in quantum wells form quite an appealing quasiparticle showing a large range of optical properties that have proven at the same time technologically useful, and physically interesting [1]. A large part of this interest is linked with the appearance of excitonic resonances in absorption up to room temperature. It is also well known, since the seminal work of Weisbuch et al [2], that free excitons appear to dominate the luminescence response of semiconductor quantum wells at low temperatures. Part of the origin of this effect lies in the breakdown of the translational symmetry which brings a very efficient recombination channel to free excitons in quantum wells [3,4].Interestingly, in the low density regime, time resolved luminescence (TR-PL) in quantum wells is observed to be always dominated by light coming at the exciton energy, even under non resonant excitation. The observations of this dominant contribution are so numerous that only a very partial list of references may be given here [5,6,7,8,9] (in order to be specific, we only consider here the case of quantum wells grown on GaAs substrates). A double question has then been debated for more than 10 years in the literature: first how do free electron hole pairs bind into excitons and second does indeed the luminescence at very short time proceed from bound excitons. A brief survey of the literature allows to find that experimentalists have reported formation times ranging from less than 10 ps up to about 1 ns [7,8,9, 10] and theoretical values range from 100 ps [11,12,13] to over 20 ns [14]. Clearly, the origin of this spreading in the reported values lies in the poor sensitivity of the experiments used in general to probe the exciton formation process, except for the case of the recent terahertz absorption experiments [10]. On the theoretical side, binding of an electron hole pair into an exciton requires, at low temperatures, the emission of an acoustic phonon, which brings long formation time due to the small coupling of acoustic phonons to excitons.The long formation time of excitons, together with the observation of luminescence at the exciton energy at the shortest times [7,15] led Kira et al [16] to introduce the idea that a free electron hole plasma, properly including Coulomb correla...