Beliefs and expectations, also referred to as priors, shape how we perceive our environment. In a complex and ever-changing world, prior beliefs need to be flexibly and continuously integrated into sensory decision processes to guide adaptive behavior. Despite the advances in understanding the integration of priors in visual decisions at the computational and algorithmic levels, its neural mechanisms remain unclear. The goal of this study was to understand the neural dynamics involved in belief formation and updating. Here, we used high-density electroencephalogram recordings to constrain and temporally dissociate prior beliefs from the visual input to investigate the neural correlates of flexible integration of prior beliefs in visual decisions. We established a novel behavioral paradigm that combines reversal learning with a difficult motion discrimination task at perceptual threshold. This task enabled the dissociation between learning of the prior beliefs, defined as a subjective probability, from the visual decision process. Behavioral choice was modeled using Hierarchical Gaussian Filtering to estimate the prior belief of motion direction. We show that single-trial factors like motion direction and choice, as well as prior and posterior belief strengths, can be decoded above chance from event-related potentials. We compare the transformation of the prior belief when motion is congruent or incongruent with the prior belief until a motor decision is made. These results elucidate the neurophysiological basis of the influence of priors during evidence accumulation of perceptual stimuli during uncertainty. Future analyses will investigate if these dynamics account for individual differences in learning under volatile environments.