Purpose: Vocabulary skills in children are typically measured with norm-referenced assessments of receptive and expressive vocabulary. Language sample analysis is an alternative method of examining vocabulary actually produced in communicative events and may be better suited to exposing subtle vocabulary weaknesses. Here, we examine the relationship between norm-referenced vocabulary testing and language sample analysis in preschool children, both children with typical hearing (CTH) and children who are deaf or hard of hearing (DHH). Method: Expressive and receptive vocabulary and spontaneous language samples were collected. Language samples were analyzed for complexity (mean length of utterance in words [MLUw]) and variability (number of different noun and verb types produced). Results: CTH had significantly higher scores on both expressive and receptive norm-referenced tests and produced sentences with greater syntactic complexity and semantic variability. Relationships between expressive test scores and MLUw were seen in both groups; the number of noun/verb types produced was related for children who are DHH only. Receptive vocabulary was not related to spontaneous spoken language for CTH. Receptive vocabulary was significantly related to MLUw and noun/verb types for children who are DHH. However, when the DHH group was subdivided into performance above and performance below the 50th percentile, relationships held only for the group with below-average performance. Conclusions: Results suggest that single-word vocabulary, norm-referenced measures indicating performance above the 50th percentile may not be sufficient to capture nuanced difficulties with vocabulary in children who are DHH. For children who are DHH, performance in the "range of normal" on a norm-referenced test may not capture ways in which their language performance differs from that of peers with typical hearing.