Five experiments examined the organization of spatial memory of an irregular path learned by walking with vision. Two hypotheses were tested: (a) that participants would establish a single global frame of reference to organize the spatial memory of the path; or alternatively (b) that participants would establish path-aligned reference directions at each path leg but not establish a global frame of reference. Participants donned a head-mounted display and were asked to navigate through a virtual six-segment path with each turning point indicated by a virtual object. The six legs consisted of two groups of three legs. The legs within groups were aligned (parallel or orthogonal) with each other and between groups were misaligned (45° tilted) with each other. At each leg, participants only perceived the object at the end of this leg. After participants walked the legs 10 times they conducted judgments of relative directions (JRD, e.g. âimagine you are standing at X, facing Y, please point to Zâ). The imagined headings in JRD were parallel to the experienced path legs. The paths varied in terms of the salience of the longest leg. Appearance of a room was also manipulated to highlight one group of the legs. The results showed that participants demonstrated significantly lower pointing error for (a) the longest leg when there was no room or (b) the first walking leg when there was no obvious longest leg or the longest leg was misaligned with the room. Pointing error was equivalent for the longest leg and the first walking leg when the longest leg was salient and misaligned with the first walking leg. These results suggested that participants established a single global frame of reference when there was a single salient context cue. However, two oblique reference frames can be established when there are two inconsistent contextual cues.