Articulatory generalization of velar cognates /k/, /g/ in two phonologically disordered children was studied over time as a function of sequential word-morpheme position training. Although patterns of contextual acquisition differed, correct responses to the word-medial, inflected context (e.g., "picking," "hugging") occurred earlier and exceeded those to the word-medial, noninflected context (e.g., "bacon," "wagon"). This finding indicates that the common view of the word-medial position as a unitary concept is an oversimplification. Possible explanations for superior generalization to the word-medial, inflected position are discussed in terms of coarticulation, perceptual salience, and the representational integrity of the word.