Purpose: Children's ability to adjust one's language according to discourse context is important for success in academic settings. This study examined whether second graders vary in linguistic and discourse features depending on discourse contexts, that is, when describing pictures in contextualized (describing the picture to an examiner while looking at it together) and decontextualized (pretending to describe the picture to a friend while sitting in front of the examiner) conditions. Method: A total of 330 English-speaking second graders in the United States (M[subscript age] = 7.33 years; 53% boys; 55% Caucasian children, 35% African American children) described three pictures in contextualized and decontextualized conditions. Children's picture descriptions were transcribed verbatim and coded for linguistic (e.g., elaborated noun phrase) and discourse (e.g., proper character introduction, degree of decontextualization) features. Results: Type-token ratio was higher in the contextualized condition than in the decontextualized condition, whereas certain types of elaborated noun phrases (e.g., simple descriptive noun phrase, noun phrase with postmodification), coordinating conjunctions, and nonclauses occurred more frequently in the decontextualized condition, controlling for total productivity and student demographics. The proportion of proper character introduction was higher in the decontextualized condition, whereas higher degrees of decontextualization and complex perspective-taking were found in the contextualized condition. Conclusion: Various linguistic and discourse cues illustrated the extent to which primary grade students employ their discourse knowledge when producing oral language. [This is the online version of an article published in "Language, Speech, and Hearing Services in Schools."]