Examples of bi-level quantised images, such as the well-known dalmatian, Dallenbach's cow, or "Mooney-type" images have been used to demonstrate the intriguing case of one-shot perceptual learning. These images are meaningless when looked at for the first time, and become easily recognisable when a cue (an original, or undegraded version of the same image) is shown. Although the neural mechanisms underlying this rapid perceptual learning are not well understood, there is indication for the involvement of face and object recognition related inferotemporal regions, as well as for the involvement of attention-related parietal regions. Here we report that 4-5 year old children are unable to rely on the cue of an undegraded version of a portrait drawing when asked to identify the bi-level quantised version. This surprising inability is there even when the cue and the degraded image are simultaneously present. In order to obtain a more sensitive index of recognition in this task, we used gaze position data. Eye-movement based attentional landscapes reveal that the initially scattered attentional foci become fully organized around 2-3 main image points after cue presentation in adult observers. However, there is no change in the pattern of eye-movements after cue presentation in 4-5 year old children. Adult-like gaze pattern is observed after cue presentation in 6-year-olds even in cases where verbal reports do not indicate recognition. It remains to be seen whether 4-5 year old children are unable to employ the cue because of the immaturity of low-level spatial integration, or because of the immaturity of top-down processes at that age.