The trace-deletion hypothesis (Grodzinsky, 1990) holds that the comprehension deficit apparent in most agrammatic aphasics results from the absence of traces at the level of S-structure. This paper reports a test of this hypothesis in a case study of an agrammatic aphasic. Two experiments--one using a sentence-picture matching task, one using the truth-value judgment task-examined the comprehension of the matrix clause in center-embedded relatives such as, The tiger that chased the lion is big. These structures provide a crucial test of the trace-deletion hypothesis because comprehension of the matrix clause (i.e., knowing that the tiger is big and not the lion) is predicted to be unimpaired. Contrary to this prediction, however, the results of the present work show that comprehension of the matrix clause in such sentences is significantly impaired. We argue that a revised version of the trace-deletion hypothesis proposed by Hickok (1992a,b) can explain the present data and other previously unaccountable findings.