The spoken language skills of 20 elementary students with verified learning disabilities were explored using a story format and discourse analysis procedure. The purpose was to compare the performance of LD students on three types of story tasks (story generation from a picture, story generation from memory, and story generation based on a game) over a one-year period. Results indicated statistically significant main effects for approximately half (9/17) of the discourse variables by task and one main effect for time. Overall, stories from memory were longer and characterized by more as well as different types of errors than were stories from pictures or stories based on games. Implications for use of stories and type of story collection task in language sampling are discussed.