Previous research has resulted in inconsistent findings regarding the predictability of schizophrenic speech samples. It was hypothesized that the predictability of schizophrenic speech varies as a function of clinically manifest thought disorder. In an experiment based on the Cloze procedure, raters were asked to predict ten passages of schizophrenic speech and eight passages of normal speech under conditions of fourth- and fifth-word deletion. Differences emerged between the samples only for the fifth-word deletion procedure. When the schizophrenic samples were grouped according to the presence of thought disorder, thought-disordered speech was significantly less predictable than normal and non-thought disordered schizophrenic speech. Furthermore, non-thought-disordered schizophrenic speech was no less predictable than normal speech. It is concluded that schizophrenia should be more carefully defined and that thought disorder should be routinely assessed in future investigations.