Two large samples were assessed twice in Grade 1: the control group (Cont; N = 2302), which received no specific interventions, and the experimental group (Exp; N = 484), which received interventions associated with code-related skills (decoding and fluency). We estimated that a mean score of around 50 correctly read words per minute corresponded to a mean reading comprehension (RC) score. We then examined the relationships between decoding, fluency, RC and their associated skills, i.e. phonemic segmentation, vocabulary and listening comprehension. The causal nature of the links between decoding, fluency and RC was examined in two ways: hypothesized and experimental. Structural modeling showed that fluency was a good mediator variable between decoding and RC and explained RC. When we compared the Cont and Exp groups, the intervention had a positive effect on RC, even though linguistic comprehension was not stimulated. We conclude that fluency should be stimulated once decoding is efficient.