Three experiments tested how structural information affects the processing of fragmented forms, using evidence from both normal subjects and an agnosic patient. The stimuli were either (a) outline drawings of objects; (b) "well" or "poorly" structured fragmented forms derived from outline drawings (Experiments 1 and 2) or (c) "well" or "poorly" structured fragmented forms derived from outline drawings of meaningless forms (Experiment 3). "Well" and "poorly" structured forms varied on the properties of collinearity and closure of the elements. Subjects decided if two simultaneously presented forms had the same orientation or were mirror-reversed. Three levels of discriminability were examined: (a) for "symmetrical" forms the decision had to be based on the detection of a small feature appearing on the same or the opposite side in the two forms; (b) for "asymmetrical" forms the decision had to be based on a large part located on the same or the opposite side of each object; and (c) for "oblique" forms, matching could be based on the global orientation of the shapes. Normal subjects performed equivalently on outline drawings and "well" structured fragmented forms and worse on "poorly" structured forms. This effect was strongest on "asymmetrical" stimuli, both with meaningful (Experiments 1 and 2) and meaningless forms (Experiment 3). This indicates that the segmentation of forms into parts, for judgements with "asymmetrical" stimuli, is affected by the structural properties of collinearity and closure. For the agnosic patient, performance was similar for "well" and "poorly" structured forms, and there was better performance in the "oblique" condition. This suggests that the patient could use global shape information, but that this coexists with a selective deficit in computing local perceptual structure, based on collinearity and closure between form elements. The implications of the results for understanding form perception and the clinical use of fragmented forms, are discussed.