We measure the temporal pair correlation function g (2) (τ ) of a trapped gas of bosons above and below the critical temperature for Bose-Einstein condensation. The measurement is performed in situ using a local, time-resolved single-atom sensitive probing technique. Third and fourth order correlation functions are also extracted. We develop a theoretical model and compare it with our experimental data, finding good quantitative agreement and highlighting the role of interactions. Our results promote temporal correlations as new observables to study the dynamics of ultracold quantum gases. The intriguing effect of particle bunching was first observed in a seminal experiment by Hanbury Brown and Twiss (HBT), where they studied correlations among pairs of photons coming from a chaotic source [1]. The result, which could be explained in terms of classical waves, had a difficult route to be accepted under a particle perspective. The full quantum theory, due to Glauber [2], signed the birth of quantum optics and made the formalism, with all its physical content, available for massive particles. Over the years, analogous HBT experiments were performed with electrons [3], neutrons [4] and cold atoms [5,6]. The possibility to extract information about the quantum statistics and the coherence, paved by the HBT experience, conjugated with the capabilities of deriving the temperature and the spatial order [7], contributed to make this technique one of the most powerful to probe atomic systems. The quest to understand the behavior of more and more complex samples suggests its application to the study of strongly interacting 1D gases [8,9], disordered [10] and supersolid phases [11] and to identify non trivial excitations [12]. While first order correlations are often accessible via interference experiments, higher order correlations require in general the recording of density or atom number fluctuations by a probe sensitive enough to detect single particles (counting techniques) or, at least, atomic shot-noise (absorption imaging). In order to have a good statistical description, an average over many realizations of the system (in theory all possible realizations) is needed. Consequently correlations, especially at orders higher than two, are usually difficult to measure because of the huge statistics required for a reliable signal. Only in some limited cases, intrinsic processes in a quantum gas such as photoassociation or three body losses can be used as a sensitive probe for higher order correlations at zero distance [13,14]. The direct observation of third order correlations is still challenging and, using standard techniques, requires a considerable effort in data collection and analysis [15]. In this, like in the great majority of the above mentioned experiments with ultracold gases, correlations have only been studied in the spatial domain whereas the temporal counterpart has been very poorly explored, limited only to the characterization of atomic beams [5,16]. Boosted by recent achievements [17][18][19], spatially ...