Recent advances in sensor technology promote using large sensor networks to efficiently andeconomically monitor, identify and quantify damage in structures. In structural healthmonitoring (SHM) systems, the effectiveness and reliability of the sensor network arecrucial to determine the optimal number and locations of sensors in SHM systems. Here, wesuggest a probabilistic approach for identifying the optimal number and locations of sensorsfor SHM. We demonstrate a methodology to establish the probability distribution functionthat identifies the optimal sensor locations such that damage detection is enhanced. Theapproach is based on using the weights of a neural network trained from simulations usinga priori knowledge about damage locations and damage severities to generate anormalized probability distribution function for optimal sensor allocation. We alsodemonstrate that the optimal sensor network can be related to the highest probability ofdetection (POD). The redundancy of the proposed sensor network is examinedusing a ‘leave one sensor out’ analysis. A prestressed concrete bridge is selectedas a case study to demonstrate the effectiveness of the proposed method. Theresults show that the proposed approach can provide a robust design for sensornetworks that are more efficient than a uniform distribution of sensors on a structure.