We reexamine the empirical record of the comprehension abilities of Broca's aphasic patients. We establish clear, commonly accepted, selection criteria and obtain a pool of results. We then subject these results to a detailed statistical analysis and show that these patients comprehend certain canonical sentences (actives, subject relatives, and clefts with agentive predicates) at above-chance levels, whereas comprehension of sentences that contain deviations from canonicity (passives, object-gap relatives, and clefts) is distinct and is at chance. That the latter is the case, and patients indeed guess at such structures, we show by comparing the distribution of individual results in passive comprehension to that of a model for such guessing - an analogous series of tosses of an unbiased coin. The two distributions are virtually identical. We conclude that the group's performance is stable, and well-delineated, despite intersubject variation whose source is now identified. This means that certain comprehension tests may not always be used for the diagnosis of individual patients, but they do characterize the group. It also means that group studies are not just a valid option in neuropsychology; they are a must, since demonstrations like ours indiciate very clearly that single-case studies may be misleading. As we show, the findings from any one patient, without the context of a group, may give a distorted picture of the pathological reality. Our conclusions thus promote studies of groups of brain-damaged patients as a central tool for the investigation of brain/behavior relations.