Models of evidence integration (EI) assume that the accumulation of external information alone is the dominant process during perceptual decision making until an overt response is made. In contrast, probabilistic sampling (PS) theories of the representation of uncertainty (Fiser at al. 2010) posit that time during perceptual decision making is primarily used for collecting samples from essentially static, internally represented distributions — at least for briefly presented simple stimuli. While EI predicts a decreasing trend in the correlation between participants’ estimation error and uncertainty over a trial, PS predicts an increasing trend even long after error and uncertainty reach asymptotic values. The predictions of PS have recently been confirmed in experiments using static stimuli (Popovic et al. 2013). However, the precise relationship between EI and PS in the more general case of dynamic stimuli have remained unexplored. To dissect the contributions of EI and PS to perceptual decisions, we used a variant of the classical random dot motion task that required estimation (rather than discrimination judgment. In each trial, participants reported their best estimate for stimulus direction and their subjective uncertainty about it by the direction and length of a line drawn on a tablet. We controlled EI by varying the coherence of the signal (providing more or less evidence), and PS by varying the stimulus presentation time (allowing for the collection of more or less samples). In each participant, we found a marked decrease in error-uncertainty correlation in the first part of the trial, indicating EI, and a significant increase in the second part, indicating PS. Moreover, the transition between these segments shifted in accordance with the change in signal coherence. These results suggest that EI and PS during decision making work in parallel with EI taking the lead early but PS determining the later part of the process.