The improvement in performance with increasing number of frames in a random-dot kinematogram (temporal recruitment) was assessed by measuring threshold signal-to-noise ratios of direction discrimination. At fast frame presentation rates (50 Hz) thresholds fell sharply as the number of frames in the sequence increased, whereas at slow frame presentation rates (20 and 10 Hz) there was a less dramatic fall in thresholds. The similarity between the results at 20 and 10 Hz suggests that the mechanism of this less dramatic rise is relatively independent of temporal factors. The recruitment effect also does not appear to be limited by a maximum spatial range. We propose that temporal recruitment may occur via two mechanisms. One involves stimulating motion detectors with greater spans and delays, whilst the other involves the co-operative interaction of signals from units tuned to similar directions and have similar spans and delays. This distinction is supported by a further experiment which eliminates the first of these recruitment mechanisms by destroying possible correlations between non-adjacent frames.