Microgrids are power systems consisting of an electrical network composed by distributed loads and generation units that may include a communication network for improved operation. The considered microgrid in islanded mode is driven by voltage source inverters implementing decentralized droop control for active power sharing together with a communicationbased consensus algorithm for frequency regulation. This paper analyses the microgrid performance subject to network failures that provoke network partitions. It is considered that the electrical partition leads to several sub-microgrids working in parallel where the power demand can be always guaranteed by the generation units, and the communication partition leads to several consensus algorithms also working in parallel. The double partitioning is analyzed through a closed-loop system model derived using the power flow equations that includes the electrical and communication connectivity. Analytical expressions for the steady-state values for both frequency and active power depending on the partitioning are derived. Selected experimental results on a low-scale laboratory microgrid illustrate the (undesirable) impact that unexpected partitions have in system performance.power sharing together with a communication-based consensus algorithm for frequency regulation [6].Apart from the known impact that natural disasters have in power systems, e.g. [7], the use of communication networks also impacts power systems, e.g. [8]. Hence, the interdependency between the electrical network and the communication network makes the problems of reliability, operation, and security more complex than in the traditional power grid.