This paper addresses the problem of scheduling inpatient admission in a hospital with highly uncertain length of stay and with a significant part of patients from emergency department. The main difficulty is to keep enough beds for unknown emergency patients and unknown future inpatients, also called elective patients, when planning admission of elective patients. For this purpose, we model inpatient admission scheduling as a stochastic programming problem. We propose an average sampling technique to estimate the number of beds needed for emergency patients and unknown inpatients. Three strategies are proposed to solve the stochastic programming problem. Experiments with data sets derived from data collected from a French medium-sized hospital are conducted to assess the performance of the three strategies.