The present study examined whether explicit training of letter-clusters leads to more gains in word-reading speed than training of the separate letters of the same clusters. Ninety-nine poor reading second-grade children were randomly assigned to a cluster-training, a parallel letter-training, or a no-training condition. The cluster-training condition showed superior short-term and long-term improvement on rapid naming of trained and untrained letter clusters, whereas the letter-training condition showed superior short-term improvement on rapid naming of trained letters. In addition, compared with the no-training condition, the cluster and letter training showed the same superior short-term and long-term improvement on trained words and pseudowords. However, both training conditions showed only marginally more short-term improvement for untrained pseudowords and only marginally more long-term improvement on a word reading-fluency task. Apparently, improvement of rapid naming of letter clusters does not, or barely, result in improvement of untrained words and pseudowords. (Contains 4 tables.)