The dorsolateral geniculate nucleus (dLGN) of the thalamus is an essential processing stage for retinal signals to reach the primary visual cortex (V1). How these feedforward signals are modulated by corticothalamic (CT) feedback and behavior remains an open question. To quantify how feedforward, feedback and behavior combine in shaping dLGN responses, direct and selective control of CT feedback in behaving animals is needed, as well as a computational model that takes the various contributing factors into account. We here recorded extracellular responses in dLGN of awake mice to a rich, dynamic movie stimulus, while we selectively and reversibly photo-suppressed V1 layer 6 CT pyramidal cells and simultaneously tracked locomotion behavior and pupil size. We predicted dLGN responses using a Linear-Nonlinear-Poisson (LNP) model with a spline basis (RFEst toolbox [1]) to estimate spatiotemporal receptive fields (RFs) and kernels for CT feedback and behavioral modulations. We found that the spline-LNP model successfully captured diverse spatial and temporal RF shapes, such as different RF polarities and uni- vs. bimodal temporal responses. The shapes of the modulatory kernels allowed to independently quantify their contributions: we found, on average, positive kernels for running and pupil size, consistent with the overall enhancement of dLGN responses with behavioral state; we also found, on average, negative kernels for optogenetic feedback suppression, capturing the removal of top-down excitation. Interestingly, effects of CT feedback varied between spontaneous and stimulus-driven periods, and between different contrasts within RFs. Finally, training models on either movies or artificial noise stimuli revealed RFs with similar characteristics, although the noise stimulus elicited overall lower firing rates. By integrating feedforward drive, feedback modulation, and behavior into an interpretable spline-LNP model for dLGN activity, this work presents an important step towards a quantitative understanding of how responses to complex, naturalistic stimuli are modulated by CT feedback and behavior.