In 4 naming experiments we investigated how Italian readers assign stress to pseudowords. We assessed whether participants assign stress following distributional information such as stress neighborhood (the proportion and number of existent words sharing orthographic ending and stress pattern) and whether such distributional information affects naming speed. Experiments 1 and 2 tested how readers assign stress to pseudowords. The results showed that participants assign stress on the basis of the pseudowords' stress neighborhood, but only when this orthographic/phonological information is widely represented in the lexicon. Experiments 3 and 4 tested the naming speed of pseudowords with different stress patterns. Participants were faster in reading pseudowords with antepenultimate than with penultimate stress. The effect was not driven by distributional information, but it was related to the stage of articulation planning. Overall, the experiments showed that, under certain conditions, readers assign stress using orthographic/phonological distributional information. However, the distributional information does not speed up pseudoword naming, which is affected by stress computation at the level of the articulation planning of the stimulus. It is claimed that models of reading aloud and speech production should be merged at the level of phonological encoding, when segmental and metrical information are assembled and articulation is planned. (Contains 8 tables, 4 figures and 10 footnotes.)