In this paper, we use experimental data collected from an Austin, Texas smart grid test bed with a system-level battery energy storage model to assess the lifetime of batteries in a microgrid operating in islanded mode during a distribution-level outage. We consider a hypothetical microgrid consisting of 21 single-family detached homes and three transformer-level community energy storage (CES) battery units ranging in size from 25 kW h to 75 kW h. To describe the performance of CES batteries, we implement a dynamic behavioral circuit model capable of describing voltage transients and rate-capacity effects. We use one-minute electricity production and consumption data collected from the smart grid test bed in 2012 to assess how the timing of an electric outage affects the islanding lifetime of a residential microgrid. We contrast our results with the average outage duration reported by U.S. electric utilities to quantify how often a residential microgrid could withstand a typical outage. Our results show that increasing the amount of rooftop PV in a residential microgrid does not significantly increase how often it can withstand an average-duration outage. However, combining PV with CES extends the median islanding lifetime by up to 11.6 h during morning outages. Based on our results, 50 kW h CES provides the best tradeoff between the cost of a CES system and its reliability benefit, allowing downstream loads to withstand an average-duration outage approximately 93% of the time. (C) 2014 Elsevier Ltd. All rights reserved.