It is well established that phonological awareness (PA) and rapid automatized naming (RAN) tasks reliably predict children's developing word reading abilities across a wide range of languages. However, existing research has not yet demonstrated unequivocally whether RAN and PA are independently and causally linked to reading, nor has it fully explored the underlying cognitive mechanisms. Most existing research has assessed PA and RAN in children who may already have some reading skill, making direction of influence hard to ascertain. To address this, the current longitudinal research initially assessed RAN and PA in a very young sample of 91 English children (M age = 3 years 11 months; SD = 3.7 months) demonstrated to be nonreaders. Children were reassessed on RAN, PA, and word-level reading at 18 months (Time 2) and then a further year later (Time 3). To explore underlying mechanisms, separate measures of reading accuracy and fluency were taken, and reading tasks varied according to the extent to which they required alphabetic decoding and lexical, orthographic knowledge. Path analyses revealed that from Time 1 to Time 2, both RAN and PA predicted word reading, indicating temporal precedence, though there was some degree of reciprocity in these relationships. However, by Time 3, while RAN still predicted accuracy and fluency of reading, PA only predicted reading accuracy. Furthermore, findings suggested that while RAN was robustly related to both alphabetic decoding and lexical, orthographic aspects of reading, PA's relationship was restricted to alphabetic decoding accuracy. Theoretical and practical implications are discussed.