This paper presents a methodology for the design of optimal load shedding against transient instability of power systems. The proposed method is considered as a security constrained, multi dimensional optimization problem which is solved by Big Bang-Big Crunch (BBBC) method. The objective function to be minimized is the total weighted load shedding and the constraints of the optimization are based on stable operation of the power system after a critical contingency occurs and on the shedding limits. During the optimization, the stability statuses of potential solutions are determined by a Probabilistic Neural Network (PNN) instead of using time domain simulations. To investigate the effectiveness of the method, imulations are carried out on the IEEE 163 bus, 17 generator test system and their results are discussed.