The perspective-taking task is one of the most common paradigms used to study the nature of spatial memory, and better performance for certain orientations is generally interpreted as evidence of spatial representations using these reference directions. However, performance advantages can also result from the relative ease in certain transformations/rotations. To differentiate spatial memory from spatial transformations, the present study took a new approach based on the hypothesis that responses may be biased toward the original representation but not a transformed one. Participants memorized a regular target array and then judged the relative direction between 2 targets while imagining facing various directions. Their response time and absolute errors showed the standard advantages at 4 imagined orientations. In contrast, an attraction analysis suggested that only 1 orientation was represented in memory, whereas performance advantages at other orthogonal orientations were due to lower transformation costs and should not be interpreted as spatial representations. These findings challenged the traditional performance-based interpretations of perspective change tasks and provided a new research paradigm to differentiate spatial representations from spatial transformations.