The effectiveness of a model for training word recognition was tested with 60 eeucable mentally retarded children (mean age 10 years). Thirty Ss, randomly assigned to the experimental group, were given 14 weeks of training on the model's seven component skills (such as the ability to construct a word given an initial sound), while the 30 control Ss were given reading tasks not related to the model. Posttest scores revealed the experimental group was significantly superior on five of the seven components of the model when a reaction time measure was used, and on six of the seven components when a number of correct responses measure was used. The experimental group was also superior on latencies for recognition of known words, on recognition of unknown words in context, and on a cloze test. Results indicated that components of word recognition can be trained in mentally retarded children and that this training transfers to better word recognition performance on a number of indices. (Experimental data and the test of component skills are appended.) (LS)