When conducting classroom observations, researchers are often confronted with the decision of whether to conduct observations live or by using pre-recorded video. The present study focuses on comparing and contrasting observations of live and video administrations of the Classroom Assessment Scoring System-PreK (CLASS-PreK). Associations between versions, mean differences, reliability, and predictive validity were examined. Results generally indicated high correlations between versions. Video codes were slightly lower on average than live codes. Reliability was generally acceptable in terms of Cronbach's alpha, but multigroup confirmatory factor models suggested some differences between observation types. Finally, CLASS scores based on each observation type indicated some predictive validity of children's academic achievement, but no observation type was uniformly better. The discussion focuses on why the codes might differ and the implications of those differences. (author abstract)