Background: When reading a word, skilled adult readers automatically decompose the word into its separate morphemes by processing the word's morpho-orthography. In children, however, it still remains unclear when and how they start to automatically decompose words into morphemes. Methods: To better understand how primary school children learn and integrate automatic morphological processes into their reading, we conducted a masked priming experiment with n = 218 first to fourth graders and a control group of 36 adult readers. Participants saw prime words on a computer screen for 67 ms, followed by a lexical decision task. For each target word, we constructed three prime words: an unrelated control prime, a semantic control prime and a test prime. The test prime was either morpho-semantically, morpho-orthographically or purely orthographically related to the target word. We analysed error rates and reaction times with linear mixed-effects models and linear combinations. Results: The error analysis revealed one significant interaction in the morpho-semantic condition for fourth graders. The reaction time analysis revealed different priming effects depending on age group: first and second graders showed no priming effects, while third graders showed priming in the morpho-semantic condition, and fourth graders and adults showed priming both in the morpho-semantic and morpho-orthographic condition. Conclusions: We conclude that (i) fourth graders automatically decompose written words into morphemes based on the words' morpho-orthography, (ii) third graders automatically detect stems in morphologically complex forms but do not rely on morpho-orthographic representations while (iii) first and second graders show no indication of either of these processes. Our results are theoretically consistent with the 'semantic view' within a localist account and can be explained by edge-aligned embedded word activation.