The ability to remember spatial locations is critical to human functioning, both in an evolutionary and an everyday sense. And yet, spatial memories and judgments often show systematic errors. Explanations for such errors have ranged from assumptions that memories are nonmetric, to the use of imperfect inferences, to the optimal combination of multiple sources of information. More recently, bias has been explained through the Category Adjustment Model – a Bayesian model in which fine-grained and categorical information are optimally combined (Huttenlocher, Hedges, & Duncan, 1991). However, experiments testing this model have largely used locations contained in simple geometric shapes. Use of this paradigm raises the issue of whether the results generalize to location memory in the complex natural world, as it should if it is to provide an over-arching framework for thinking about spatial memory. Here, this issue is addressed using a novel extension of the location memory paradigm that allows for testing of location memory in an everyday, 3D environment. The results support two predictions of the Category Adjustment Model - that memory for locations is biased toward central values, and that the magnitude of error increases with the retention interval. Future directions for testing the model in an increasingly ecologically valid manner are discussed.