An inherent limitation of human visual system research stems from its reliance on highly controlled laboratory conditions. Visual processing in the real world differs substantially from such controlled conditions. In particular, during natural vision, we continuously sample the dynamic environment by variable eye movements that lead to inherent instability of the optical image. The neuronal mechanism by which human perception remains stable under these natural conditions remains unknown. Here, we examined a neural mechanism that may contribute to such stability, i.e., the extent to which neuronal responses remain invariant to oculomotor parameters and viewing conditions. To this end, we introduce an experimental paradigm in which intracranial brain activity, a video of the real-life visual scene, and free oculomotor behavior were simultaneously recorded in human patients. Our results reveal, in high-order visual areas, a remarkable level of neural invariance to the length of eye fixations and lack of evidence for a saccade-related neuronal signature. Thus, neuronal responses, while showing high selectivity to the category of visual images, manifested stable "iconic" dynamics. This property of invariance to fixation onset and duration emerged only in high-order visual representations. In early visual cortex, the fixation onset was accompanied with suppressive neural signal, and duration of neuronal responses was largely determined by the fixation times. These results uncover unique neuronal dynamics in high-order ventral stream visual areas that could play an important role in achieving perceptual stability, despite the drastic changes introduced by oculomotor behavior in real life.