Misrepresentative occupancy schedules cause uncertainty in building energy simulations which often results in an undesired gap between actual performance and predictions. To address this issue, stochastic building occupancy models, which commonly use time-use surveys (TUS) as their input, have been proposed as a replacement for the conventional generic static schedules. This paper describes the development of a two-state stochastic occupancy model based on the 2017 American TUS (ATUS) that takes account of differences between weekdays and weekends. In this model, survey respondents were clustered based on the number of residents in their households and a first-order Markov-Chain technique was used to generate occupancy presence schedules.