Descriptive assessments involve recording naturally occurring instances of behavior and corresponding antecedent and consequent events. Authors have argued for the use of two forms of descriptive assessment, structured and narrative Antecedent-Behavior-Consequence recording, because these methods may require little training. However, the extent to which minimal training produces accurate data with these methods is unknown. During Experiment 1, we examined teachers' accuracy when recording descriptive data from videos. Accuracy on problem behavior did not improve over time, regardless of initial exposure to structured or narrative ABC recording. Teachers preferred the structured ABC recording sheet. During Experiment 2, we provided training to participants using an automated procedure that included practice and feedback. Accurate data collection on problem behavior increased for six participants after training. Data-collection accuracy was higher for environmental events involving the presentation of stimuli (demand and attention) than the absence of stimuli (escape and low attention). Participants displayed idiosyncratic preferences for the structured or the narrative ABC recording sheet.