A fuzzy direction neural network classifier used for fault detection and isolation (FDI) in a liquid propellant rocket engine is proposed. The fuzzy direction neural network utilizes fuzzy sets as engine fault classes. Each fuzzy set is an aggregate of fuzzy direction bodies. A fuzzy direction body is described by a direction vector，an included angle and two radii. The fuzzy direction neural network can learn nonlinear direction boundaries in a single pass through the training data. The FDI simulation research has shown the strong discernibility of the fuzzy direction neural network.