Four equated groups of moderately retarded children were trained on a simple eigth-word reading vocabulary, each under different experimental condition. In the standard condition a printed word was repeatedly paired with a pictured representation. In the faded condition the same procedure was used except that the picture was gradually faded-out over trials. A mixed condition had , on alternate trials, either word-and-picture or word alone, while on alternate trials a control condition had either picture alone or word alone. On immediate and delayed tests of vocabulary learning, subjects showed large differences in reading performance with these methods. The order of efficiency of conditions was faded, mixed, standard, and control. Several theorical interpretations of the superiority of fading were considered. The conclusion favored was that the fading procedure was best because it increased the attention-value of the words in relation to the faded pictures, without sacrificing the superior associative effects of noncontingent pairings of words and pictures.