Discrimination learning studies suggest that horses learn more easily using spatial than visible object-specific (OS) cues. However, spatial cues have generally confounded intra-array, distal and/or egocentric spatial information. It is also unclear whether conflicting cues compete for association or are redundantly encoded, and furthermore, the influence of prior experiences or training has not been quantified so far. We examined the effect of cue modality on unweaned foals' performance in a discrimination learning task. After a pilot study confirmed that horses could perform the required OS cue discrimination, nine foals learnt to find food in one of three covered buckets, in any of four positions within a test arena. In Stage 1 the rewarded bucket was signified both by OS cues (pattern) and by relative spatial cues (position). On reaching criterion, cues were separated and foals had to ignore the inappropriate cue (Stage 2). Foals assigned to follow spatial cues (n = 5) completed Stage 2 faster than foals for whom OS cues remained consistent (n = 4). Spatial group foals all reached criterion without delay; no foal in the OS group reached criterion within the testing period. OS group foals initially persisted in responding to the previously correct position, adopting spatially-based strategies when this proved unsuccessful. The findings show for the first time that, even in the absence of distal spatial information, intra-array spatial cues were more salient than OS cues for foals in a food-finding task and learning appeared rather inflexible.