Perceptual decision-making is thought to involve a gradual accrual of noisy evidence. Temporal integration of the evidence reduces the relative contribution of dynamic internal noise to the decision variable, thereby boosting its signal-to-noise ratio. We aimed to estimate the internal evidence guiding perceptual decisions over time, using a novel combination of external noise and the response signal methods. Observers performed orientation discrimination of patterns presented in external noise. We varied the contrast of the patterns and the delay at which observers were forced to signal their decision. Each test stimulus (patterns and noise sample) was presented twice. Across two experiments we varied the availability of the visual stimulus for processing. Observer model analyses of discrimination accuracy and response consistency to two passes of the same stimulus, suggested that there was very little growth in the internal evidence. The improvement in accuracy over time characterised by the speed-accuracy trade-off function predominantly reflected a decreasing proportion of non-visual decisions, or pure guesses. There was no advantage to having the visual patterns visible for longer than 80 ms, indicating that only the visual information in a short window after display onset was used to drive the decisions. The remarkable constancy of the internal evidence over time suggests that temporal integration of the sensory information was very limited. Alternatively, more extended integration of the evidence from memory could have taken place, provided that the dominant source of internal noise limiting performance occurs between-trials, which cannot be reduced by prolonged evidence integration. (Contains 2 tables and 11 figures.)