Intra pulse feature extraction is a key issue in advanced radar emitter signal recognition. A novel approach was proposed to extract intra pulse features from radar emitter signals to recognize different radar emitters. Fractal dimensions of radar emitter pulse signals were regarded as the classification features to recognize different intra pulse modulations. With the change and distribution information of magnitudes, frequencies and phases of radar emitter signals, the features reflect the laws of intra pulse modulation of radar emitter signals. It is proved in theory that the fractal features are not sensitive to noise, which is also shown in simulation experimental results. Experimental results of fractal feature extraction and automatic recognition of 10 typical radar emitter signals show that the fractal features have large between classes distance and small within class distance, and they achieve high accurate recognition rate, which verifies the approach is valid.