No direct information is available on the factors which enter into an operator's decision as to whether two moving objects will collide, nor are there data on the accuracy with which such predictions are made. The previous studies which bear the closest relation to these problems have involved only one moving object. In a motion-picture Estimation of Velocity Test devised by Gibson, S was required to estimate at a subsequent instant of time the position of a moving airplane which had disappeared behind a cloud.1 The responses indicated that S generally was able to extrapolate the target-motion with an error of velocity of less than 20%. Experimental work by one of the present writers, using a task in which S was required to continue tracking after the target disappeared, found the average error of continuative velocity to be from 12 to 17% of the previous velocity.2 The continuation of accelerated patterns of motion also was studied. It was found that the continuation maintained a constant velocity which was lower than the last visible velocity for positively accelerated targets and higher for negatively accelerated ones. It was inferred that S had either used a particular 'stale' velocity or else had integrated velocities over some period of time. In the present experiment, one type of collision-situation was studied. Two targets in marked orthogonal lanes approached an intersection, and a centrallyplaced S predicted their future relative positions. There are, of course, many important collision-situations to which one or more of these specifications will not apply, and it should be emphasized that each variation in situation may well bring new factors into the judgments. Expectations here were based upon the earlier studies on estimation and continuation. It was predicted that judgments of the relative future positions of two moving targets would reflect an error of 15 or 20% in estimating the velocities. The smoothing of accelerated motion also was foreseen. METHOD