Purpose: This study investigated how correlations between rapid automatized naming (RAN) and reading depend on characteristics of the stimuli. RAN tasks using stimuli with high phonological demands were predicted to be the strongest correlates of decoding efficiency, while high semantic demands were predicted to lead to stronger correlations with comprehension. Method: At two time points, 132 Grade 2 children completed four different RAN versions, two using letter stimuli (low semantic load) and two using object stimuli (high semantic load). Both types of stimuli were used in either a repeated version, where a set of four items were repeated multiple times (low semantic load), or in a unique version, where each item appeared only once (high semantic load). Decoding efficiency and reading comprehension were assessed in Grade 5. Results: Analyses showed that confirmatory factor models with separate factors for each version provided better fit than grouping factors according to time point. Repetition (lowering semantic load) increased the longitudinal association between RAN objects and decoding efficiency. There was a tendency for conditions with higher semantic load to correlate more strongly with reading comprehension after control for decoding efficiency, but the differences were not significant. Conclusion: The results indicate that increasing semantic load weakens the relationship with decoding efficiency.