A repertoire of polysyllabic words, produced by an approximately 2-year-old child, was analysed for primary stress application. A general rule for stress application did not emerge from the data. Instead, much variability existed among tokens of individual lexical items (which were imitative and/or spontaneous) and among different lexical items with the same stress contour, regarding the placement of primary stress. Intraword and interword variation suggested the existence of four groups of words treated differently with respect to primary stress placement. The word groups were separated by their characteristic stress contours, which were (1) conventional primary stress, (2) misplaced stress, (3) level stress, and (4) undetermined stress. In general, consistent application of conventional stress was related, significantly, with lexical items represented by primarily spontaneous tokens. Results support a view of lexical primacy during early stages of learning word stress.