This paper investigates the dependence of estimates of reservoir storage capacity derived using behavior analysis on the length of inflow sequence used for overyear reservoir simulation. It has generally been assumed that simulation using behavior analysis, which incorporates a given probability of failure, will always give steady state estimates of the storage capacity (apart from the effects of the initial reservoir condition). The results reported here show that it may take sequence lengths as much as 1000 years or more for the mean of the distribution of storage capacity estimates to approach a stationary value. For some cases with high draft and high inflow variability, we show that a significant swing of the mean storage estimates from an initial downward bias into an upward bias occurs before their convergence to a stationary level. However, the median storage estimates always showed downward bias which sometimes decreased very slowly with increasing sequence length. We provide explanations for these observations and discuss some of the implications on the choice of inflow sequence length in determining reservoir storage capacities using behavior analysis. ¿ 1997 American Geophysical Union