The shape of radio signals determined by modulation system and that produced by the nonlinear transform are different in kind, which is the basis for discriminating them. Feature extraction based on fractal theory is a new method. In this paper,the features of modulation and nonlinear transform are projected to fractal feature space by computing the variance fractal dimension and the Mandelbrot singularity spectra dimensions. The numerical result show that differences in fractal features do exist between different modulated signals and the nonlinear transformed signals. The results show that the feature of these signals can be extracted based on fractal theory, which has engineering value for the identification of modulated radio signals.