Purpose: Lexical stress and phoneme processes converge during phonological encoding, but the nature of the convergence has been debated. Stress patterns and phonemes may be integrated automatically and rigidly, resulting in a unified representation. Alternatively, stress and phoneme may be processed interactively based on sublexical contexts. The purpose of this study was to evaluate the extent to which the lexical stress and phoneme processing interact in a novel nonword learning paradigm. Method: Twenty-seven adults with typical speech skills were trained to produce nonwords with specific phonemes, syllables, and stress patterns (Set 1) to an accuracy criterion. Then, participants repeated nonwords that varied from Set 1 in syllable position (Set 2), phoneme sequence (Set 3), included new phonemes (Set 4), or had new phonemes and stress patterns (Set 5). Nonword productions were perceptually analyzed, and phoneme and stress errors were counted. Results: Participants' produced Set 1 nonwords with few phonemic or stress errors after training; a similar number of both types of errors were produced when comparing Sets 2 and 3. Greater phoneme and stress errors were produced on nonwords from Sets 4 and 5 compared to Sets 1-3. The highest number of phonemic errors occurred in Set 4 nonwords. There was no difference in the number of stress errors produced on nonwords in Sets 4 and 5. Conclusion: The results of this study suggested that lexical stress and phoneme processing co-occurred and interacted during nonword productions. Trained stress patterns were learned during training; however, no evidence for a unified representation was observed. Negative interference was observed in nonwords with new phonemes and trained stress patterns, suggesting online phoneme processing may have dominated and interfered with the retrieval of stored metrical frames.