A probability hypersphere set neural network was proposed for engine Fault Detection and Diagnosis (FDD). The neural network utilized probability sets as engine fault classes. Each probability set was an aggregate of some hyperspheres.A hypersphere was described by a center and a radius. The probability hypersphere set neural network can learn nonlinear failure boundaries in two passes through the training data,and provide the ability to incorporate new and refine existing failure classes without retraining. The FDD simulation of YF20 engine systems demonstrates the strong qualities of the probability hypersphere set neural network.