The present study examined whether spatial information of a novel environment was integrated within a reference frame during initial learning, or only later when required for pointing to other targets. Twenty-two participants repeatedly walked through a multicorridor virtual environment, presented via head-mounted display. At several stages within the learning process they were teleported to locations along the route and asked to self-localize and point to other locations. Pointing was faster during later tests as well as for closer targets, both of which might require less integration. In a control group, the same exposure to the environment, but without testing, did not yield this reduced pointing latency. These results are inconsistent with the assumption that participants had already integrated spatial information within a single reference frame during learning and simply accessed this information during testing. Rather, spatial integration is a time consuming process which is not necessarily undertaken if not required.