Power distribution is an essential service for almost all human activities, including industry, commerce, services, habitation. The occurrence of failures in a power network can cause negative impacts from the customer's perspective and also economic losses for the utility given that an amount of energy is not distributed and accounted for. This work has the objective of proposing methodologies to improve power distribution network reliability. Guided by this objective, the switch allocation problem is investigated under a combinatorial optimization perspective. Switches are devices that isolate failure, restraining them from propagating throughout the network. The switch allocation problem consists in finding the best locations to install a set of switches in order to minimize failures impacts in the network. In this work, the switch allocation problem is addressed considering uncertainty in the input data regarding the system failure rates and the power demand of costumers. This uncertainty is incorporated within a formulation of a two-stage stochastic problem.