The objective of this paper is twofold: (i) to present and analyze particle temperature statistics in turbulent non-isothermal fully-developed turbulent gas–solid channel flow for a large range of particle inertia in order to better understand particle heat transfer mechanisms; (ii) to examine the performance of a recent Probability Density Function (PDF) model provided by Zaichik et al. (2011) [1]. In order to achieve such objectives, a Direct Numerical Simulation (DNS) coupled with a Lagrangian Particle Tracking (LPT) was used to collect fluid and particle temperature statistics after particles reach a statistically stationary regime. A non-monotonic behavior of particle temperature statistics is observed as inertia increases. The competition between different mechanisms (filtering inertia effect, preferential concentration, production of fluctuating quantities induced by the presence of the mean velocity and/or mean temperature gradients) are responsible for such a behavior. This competition is investigated from the exact transport equations of particle temperature statistical moments, fluid statistics conditionally-averaged at particle location, and instantaneous particle distribution in the flow field. Using these data, the accuracy of a PDF model is also assessed in the second part. From this assessment, it is seen that, despite the assumptions made, the model leads to a satisfactory prediction of most of the particle temperature statistics for not too high particle inertia.