Speakers of English and German typically adopt the reflection variant of the relative frame of reference (FoR) in order to describe how nonoriented objects that are located in front of them are related to one another. Little is known, however, about how they proceed in dorsal settings, with objects located in their back. In this article, we explore the turn hypothesis, which assumes a (mental) 180° turn of the observer to face the objects, converting the dorsal into a frontal situation, so that the preferred FoR variant for frontal settings can be applied. To elicit spatial references, we used photographs that showed an observer and two objects either in the observer's visual field (frontal condition) or in the observer's back (dorsal condition). The observer was looking either in the same direction as the referencing individual (aligned perspectives) or in the opposite direction (vis-à-vis perspective). Data from two experiments show that while participants do adopt the observer's perspective, their references in dorsal settings are incompatible with the turn hypothesis. Analyses of response latencies indicate additional cognitive costs for establishing a FoR for the very first item in the dorsal condition as compared to the frontal condition, but fast adaption for subsequent items, and high intraindividual consistency in FoR choice in both conditions. Maintaining the assumption that references in dorsal settings should be compatible with the variant of the relative FoR adopted in frontal settings, participants' references can be explained by assuming a backward projection that gets by without a (mental) turn of the observer.