A new Fisher discriminant function based passive islanding detection approach is proposed for distributed generation system in this work. Through observing the harmonic components in the output instantaneous power of distributed generation system, six-dimensional features are obtained for islanding detection. Based on utilizing Fisher discriminant function, a model is constructed to convert the obtained features into one dimensional data space and the related threshold is calculated. The original data are then mapped into two categories of islanding condition and non-islanding condition. The approach is verified with simulation and experimental results.