Previous investigations of visuospatial abilities in the visual form agnosic patient D.F. suggest that her egocentric sensorimotor processing is intact while her 'allocentric' judgments of spatial position are impaired. The current investigation extends these previous observations by comparing D.F.'s performance at pointing to a set of spatially distributed stimuli, either directly or by 'pantomiming' the responses in an adjacent homologous workspace. The results showed accurate sensorimotor localization when D.F. pointed directly to single targets or to sequences of targets, presumably as she could use egocentric visual coding. In spite of making relatively spared spatial judgments about the arrays, however, D.F. performed quite poorly when copying them and on the pantomimed pointing task. In this latter task good performance presumably depends on an ability to represent both the categorical and coordinate properties of the array (as does copying them), and to translate these into the effector-based coordinates required for accurate action. D.F.'s pantomimed pointing was similar to her copies of target arrays, as in both tasks there was evidence of spared (although somewhat degraded) appreciation of the relative spatial positions of the stimuli. Remarkably, her accuracy in this allocentric task was not worsened by longer pointing sequences. It is possible that D.F.'s degraded performance reflects a relative (though not complete) preservation of categorical coding within the ventral stream, despite a loss of coordinate coding there.