This work was aimed to study if goldfish telencephalon is involved differentially in spatial and cue learning. With this purpose, animals were assigned to two learning conditions, 'spatial constancy' and 'directly cued', and their performance was recorded before and after ablation of the telencephalon. During the presurgical acquisition period, animals of both groups learned to solve the task with accuracy, and reached the goal in transfer tests, even though they were released from new starting positions and the response requirements were changed. Ablation impaired selectively the solution of the spatial constancy problem, but had no significant effects on the cued condition. However, with additional training, performance of the ablated animals in the spatial constancy condition improved to control levels. The above data suggest that fishes can implement multiple spatial learning strategies which have different neural substrata. These results are discussed in relation to the possible nature of the representation underlying each task.