Municipal storage tanks are used to balance differences in supply and demand. Tanks have traditionally been sized based on deterministic criteria for balancing, fire, and emergency storage components. In this paper a stochastic analysis method is proposed to model both the deterministic and probabilistic components of consumer demand, fire demand, and pipe failures in water distribution systems. The method estimates a number of tank reliability criteria as functions of tank capacity, which provide a site-specific way of determining the required tank capacity based on user-defined reliability criteria. The method is illustrated by applying it to a typical water supply system. It was found that the tank failure duration follows a Weibull distribution. The tank failure rate was found to be very sensitive to tank capacity and can be described with an exponential distribution. It is proposed that one failure in 10 years under seasonal peak conditions is used as a design criterion for tank sizing. In many cases this will result in substantially smaller tanks than is currently specified by design guidelines.