Perception seems to be a continuous stream and, for this reason, we often implicitly assume that perception is continuous. In fact, many models of vision rely explicitly and implicitly on continuous perception. However, continuous perception is challenged by many phenomena such as apparent motion, in which we do not perceive first a static disk and then another static disk but a smooth trajectory, favoring discrete accounts. What is the sampling rate of discrete perception? Usually, the sampling rate is determined by temporal resolution. If we cannot perceive two flashes of light presented 40ms after each other, discrete sampling cannot be faster than 40ms. However, different paradigms have shown evidence for sampling rates ranging from 3ms to 300ms challenging discrete models. To overcome these challenges, first, we propose a two-step model, in which a quasi-continuous unconscious processing stage with a high temporal resolution precedes conscious discrete perception, occurring at a much lower rate, in the range of 400ms. Second, we provide evidence for this model from a set of TMS and visual masking experiments. Finally, we compare a series of mathematical models with each other and show that one stage models, continuous or discrete, cannot explain the experimental results, thus, further favoring two stage models of perception.