The invention provides an uncertainty judgment method of a passive residual heat removal system. The uncertainty judgment method has the outstanding advantages that a new sampling method is provided and fitting precision of a fitting function in an invalidation area is improved. The sampling method is capable of increasing proportion of invalid sample points in gross sample points and improving fitting accuracy of the invalidation area. The uncertainty judgment method has another advantage that relation of input and output variables is fitted by an artificial neural network, so that limitations of the quadratic polynomial on function modes are avoided. Meanwhile, reliability probability is calculated by Monte Carlo sampling instead of a simple second-order matrix method, and error caused by secondary processing is avoided.