Hippocampal place cell populations putatively encode allocentric spatial position [1-4]. In bats and rodents, place cells exhibit heterogeneously sized and overlapping receptive (or place) fields [5-9]. These heterogeneous place fields instantiate a multiscale spatial tuning landscape in which the larger place fields of some place cells encompass the relatively more numerous narrowly tuned place fields of other cells (Fig 1a). In this work, we investigate the effectiveness of such multiscale codes for encoding allocentric spatial position. We do so by deriving closed-form expressions for how well population activity discriminates between nearby spatial locations, a measure given by the population Fisher information. We found that a multiscale place field code is no more powerful in encoding position than a place field code in which tuning width is homogeneous across all neurons (Fig 1b, g vs. h). We furthermore showed that this result holds regardless of how place field centers are distributed in the environment with respect to encoded stimuli (Fig 1e,f, i vs. j). Across neurons with similar maximum firing rates (Fig 1d), neurons with larger receptive fields expend more metabolic resources than those with narrower fields. To account for this, we explored a regime in which neurons adjust their maximum firing rate to incur the same metabolic cost, irrespective of their field size (Fig 1c). Under this regime, we found that for every multiscale code, there exists a homogeneous code with the same number of neurons that is strictly better at discriminating spatial locations (Fig 1g,h). Our results generalize to arbitrary stimulus dimensionality, beyond the 2-dimensional spatial stimulus case considered here. Our conclusions stand in conflict with recent simulation-based theoretical work which, using a measure of discriminability similar to ours, argues that multiscale place fields are the optimal tuning strategy for encoding position [10]. Using our analytic approach, we find that the conclusions reached by [10] depend strongly on implicit modeling assumptions and on the authors’ choice of measure for discriminability (Fig 1k,l). Overall, our results show that conceptualizing place cells as solely implementing an allocentric positional code fails to predict their multiscale tuning. Instead, we speculate that this structure finds its purpose in supporting more complex computations underlying navigation beyond simply implementing a positional code.