Abstract: As an important element of seismic fragility and risk analysis, the probabilistic seismic demand analysis has become a main component of the new-generation of performance-based earthquake engineering (PBEE). The key issue of probabilistic seismic demand analysis is to derive a probabilistic seismic demand model, that denotes the probabilistic relationship between earthquake intensity and seismic demand. Given a specified level of earthquake intensity, seismic demand is generally assumed to submit to a lognormal distribution. A log-line relationship is assumed between earthquake intensity and the median value of seismic demand, while the logarithm standard deviation of seismic demand is assumed invariant. In order to examine the validity of the above assumptions adopted in the probabilistic seismic demand model, four RC frame buildings of various heights and fortification levels are taken as test beds, and one hundred real ground motions are selected as inputs. The results show that the assumptions of the median and logarithm standard deviation of seismic demand are adequate for a probabilistic seismic demand model considering the balance of accuracy and efficiency. However, the conventional lognormal model for seismic demand cannot incorporate the case of collapse, thus it should be modified.