Three experiments directly investigated Vellutino's (1977) verbal-deficit hypothesis, which suggests reading difficulties of learning disabled children are attributable to deficiencies in linguistic coding. Short-term retention was studied using random shapes to minimize effects of prior verbal learning. Adopting a probe-type serial memory task, normal and LD readers were compared on recall performance after pretraining of named and unnamed stimulus conditions. Recall was markedly better for normal subjects trained with names compared with the unnamed control condition, while disabled children demonstrated inferior recall on named compared with unnamed pretraining. Experiment 1 demonstrated that verbal encoding effects were stronger for normal and deaf subjects than for disabled, especially on primacy positions. Experiment 2 suggested children s (normal and disabled) idiosyncratic verbal coding did not facilitate recall compared with names provided by the experimenter. Experiment 3 suggested that names for random shapes, regardless of age, facilitated recall for normal but not for disabled readers. Verbal codes appeared particularly important in differentiating groups at primacy positions. Consistent in all three experiments, no difference was found in recall of unnamed shapes between normal and LD readers. These experiments suggest that primary reading deficits are related to integrative linguistic deficiencies and not to deficiencies of visual memory or rehearsal.