It is difficult to avoid accepting the nu11 hypothesis when the results of an appropriately designed test prove null. The present test seems theoretically appropriate, since it has long been accepted that obtaining reports on unattended material constitutes evidence for iconic storage and it has also been accepted that the integration of dot letters provides confirmation for the theory. Our experiments have combined these two approaches, showing that the predicted effect does not occur. It seems unlikely that the experiments are methodologically inappropriate since our results, where the conditions are comparable, closely resemble those of the critic. Cohene (1974) obtained little more than 30% accuracy for bigram integration at an interstimulus-interval of 0 sec. (corrected for item guessing), while Cohene and Bechtoldc (1975) only found 70% accuracy in what corresponds to our superimposed condition. It is true that the attentional demands of the task were high, but this problem is inherent in the basic claim that iconic storage provides access to a large reserve of unattended stimuli. As for the nature of the information to be integrated, one might wish to accept Campbell and Mewhort's (1980) partition of the icon into a feature buffer and a postcategorical character buffer, assigning dot storage to the feature buffer. However, as our experiment shows no excess storage of any kind, it seems that whatever features are held prove insufficient to generate any character surplus. Even a modified, postcategorical view of the icon finds no support. Since Haber (1983) has recently argued that an icon could have no practical utility, except perhaps to explain flashlight viewing or reading in a lightning storm (p. 16), it may now be advisable to omit iconic memory from our description of information processing.